impactul crizei asupra modelelor econometrice
TRANSCRIPT
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 1/37
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 2/37
Economic Papers are written by the Staff of the Directorate-General for Economic and FinancialAffairs, or by experts working in association with them. The Papers are intended to increase awarenessof the technical work being done by staff and to seek comments and suggestions for further analysis.The views expressed are the author’s alone and do not necessarily correspond to those of the EuropeanCommission. Comments and enquiries should be addressed to:
European CommissionDirectorate-General for Economic and Financial AffairsPublicationsB-1049 Brussels
BelgiumE-mail: [email protected]
This paper exists in English only and can be downloaded from the websiteec.europa.eu/economy_finance/publications
A great deal of additional information is available on the Internet. It can be accessed through theEuropa server (ec.europa.eu)
ISBN 978-92-79-22972-5doi: 10.2765/25913
© European Union, 2012
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 3/37
Inflation forecasting and the crisis: assessingthe impact on the performance of different
forecasting models and methods
Christian Buelens*
Abstract: This paper analyses how the accuracy of euro area inflation forecasting models has
been affected by the financial and economic crisis. Its first objective is to compare the
precision of three representative groups of inflation forecasting models (rules of thumb and
benchmark models; autoregressive moving average models; autoregressive distributed lag
models) under a direct and an indirect approach, respectively. Under the former, theforecasting models contain headline inflation as their dependent variable; under the latter,
individual forecasts are generated in a first step for each of the main HICP components, and
subsequently aggregated to obtain an indirect forecast of headline inflation. The second
objective is to study how the absolute and relative forecasting performances of the models and
approaches have been impacted by the economic and financial crisis.
The paper finds that direct forecasting models selected on the basis of a penalty function
generally dominate simple rules of thumb and econometric benchmark models. The analysis
furthermore suggests that when an appropriate specification for the component-specific
models is found, indirect forecasts outperform the corresponding direct forecasts.
Nonetheless, in line with the findings from earlier studies, there are insufficient elements toassert a systematic superiority of one of the two approaches. Concerning the second objective,
the across-the-board rise in the forecast errors of all models considered, confirms that
inflation forecasting has become substantially more difficult after the onset of the crisis.
However, the deterioration of the different models has been uneven. Relative to simple
econometric models and rules of thumb, the performance of direct autoregressive distributed
lag models and of the indirect approach has improved during the crisis.
Keywords: HICP, inflation, forecasting, aggregation, model selection, model evaluation,
inflation targeting
JEL Codes: C32, C52, C53, E31, E37, E58
* E-mail: [email protected]
A first version of this paper was written when I was working at the Directorate General for Economic and
Financial Affairs of the European Commission. I thank Björn Döhring, Luis Fau, Cecilia Frale, Paul Kutos,
Francesco Montaruli and Alessandra Tucci for comments on an earlier draft of this paper. The opinionsexpressed herein and any errors are my own.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 4/37
2
1. Introduction
Economic agents base many current decisions inter alia on their expectation of the future
inflation pattern. The views held about future inflation may influence firms' price-settingbehaviour or workers' wage-demands, and thereby impact current purchasing power and
labour costs; through its effect on the real interest rate and on inflation risk premia, the
expected inflation rate furthermore influences savings and investment decisions. Inflation
forecasts and projections are also often at the heart of economic policy decision-making, as is
the case for monetary policy, which in most industrialised economies is mandated to maintain
price stability over the medium term. Decision-makers hence need to have a view on the
likely future path of inflation when taking measures that are necessary to reach their objective.
Yet, while being indispensable to many decision-making agents, forming inflation
expectations is generally both complex and costly: indeed, inflation forecasting requires an
understanding of economic relationships, econometric modelling tools, access to data and
other information.
The first objective of this paper is to assess the forecasting performance of direct and indirect
euro area inflation forecasting models. While direct inflation forecasting models contain
headline inflation as their dependent variable, indirect forecasts are obtained by aggregating
component-specific forecasts, weighted by their share in the Harmonised Index of Consumer
Prices (HICP).1
The models considered in the paper are taken from three different model
groups (rules of thumb and benchmark models; autoregressive moving average models;
autoregressive distributed lag models), which vary in terms of selection procedure and
information included. As such, they are simple but representative illustrations of the more
sophisticated models used by forecasters in practice.
The first group of models comprises "low-cost" numerical and econometric benchmark
models. As concerns headline inflation, simple numerical benchmarks (“rules of thumb”) may
in fact offer an alternative to more complex forecasting techniques, and can be adopted by any
economic agent at little cost. For example, the prevailing inflation rate or the central bank’s
inflation target may be used as potential forecasts. Simple econometric models, which are
“imposed” on the data, without following a particular selection procedure, are another rapid
and inexpensive way of forecasting inflation. Typically, however, forecasters also rely on
econometric models that are more complex and not as readily available. The second group of
models considered in this paper, consists of autoregressive moving average (ARMA) models
picked on the basis of a penalty function (information criterion). For the third group of
autoregressive distributed lag (ADL) models, the information set is extended to potentialexogenous predictors of inflation.
In a first step the models are applied to headline inflation to generate direct forecasts. All
models are subsequently applied separately to the five components to obtain the indirect
inflation forecasts. This method requires the specification and selection of one model per
component, thus adding a number of intermediate steps, which are not required in the direct
approach. While the advantages of indirect relative to direct forecasts are ambiguous from a
theoretical point of view (cf. section 2), indirect forecasts are attractive in practice, as they can
1
The five main components of headline inflation are energy, processed food, unprocessed food, non-energyindustrial goods and services. While this breakdown is standard, some authors (e.g. Bermingham and
D’Agostino (2010)) also study lower levels of disaggregation.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 5/37
3
be associated to more detailed narratives and economic arguments, and avoid inconsistencies
between headline inflation and component-specific forecasts.2
The paper’s second objective is to assess the impact of the large swings in commodity prices
observed since 2007, and of the economic and financial crisis, on inflation forecasting. As
noted below, it emerges from the literature on inflation forecasting that the performance of amodel depends inter alia on the estimation and evaluation period. At the current juncture, it is
thus particularly interesting to assess to what extent the financial and economic crisis, which
itself was preceded by a shock in global food and commodity prices, has impacted the
performance of different models and approaches in both absolute and relative terms.
Substantial forecast revisions by international institutions and professional forecasters suggest
that the traditional approaches have indeed been seriously challenged (table A.1 in the annex
illustrates this by way of example, showing different vintages of inflation forecasts made from
2006 onwards for the years 2008 and 2009). The impact of the crisis on the relative
performance of different inflation forecasting models and methods is studied by repeating the
comparative evaluation for two different sample periods ending before and after the onset of
the crisis, respectively.
The remainder of the paper is organised as follows: after a brief review of the literature in the
next section, section 3 presents the pattern of euro area inflation developments since 1996.
Section 4 introduces the benchmark models considered, as well as the algorithms that are used
to select autoregressive and ADL-models. The evaluation of the different models in terms of
their forecast accuracy during the crisis is presented in section 5. Section 6 presents the
comparative assessment for the pre-crisis period, essentially showing how the conclusions
from section 5 would have been different, if the comparison had been performed three years
earlier. Section 7 concludes.
2. Literature review
Over the past two decades many central banks around the world have adopted inflation
targeting frameworks. Their central feature is a publicly announced quantitative inflation
target, typically expressed as a single number or a range, to be met over the medium term
(Bernanke and Mishkin, 1997; Hammond, 2011). The main objective of a pre-announced
target-level is to anchor inflation expectations and notably to prevent second-round effects in
the event of a transitory shock. If credible, it may thus contribute to keeping inflation stable
and low. An inflation target also provides a measure by which a central bank’s performance
can be judged ex post by the public: the closer the observed inflation rate and the inflationtarget are to each other, the better the central bank's track record, and the more compelling is
the incentive to trust the central bank as regards the future, by using the target as a forecast ex
ante.3
Inflation forecasts represent a key element in an inflation targeting framework. As
noted by Svensson (1997), a central bank’s control over inflation is imperfect and missing the
target may be the consequence of factors outside its command. That said, a central bank
2 It is also noteworthy that many published forecasts of headline inflation are based on an indirect approach; this
is the case notably for the Eurosystem staff’s inflation projection exercise and for European Commission
forecasts.3
The incentive is likely to be even greater, if it can be shown that the inflation target has also outperformedcompeting alternatives in the past, and adopting the inflation target as an own forecast would consequently have
constituted the optimal strategy ex ante.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 6/37
4
pursuing an explicit inflation target would still set its monetary policy instrument in such a
way that its inflation forecast would coincide with the official inflation target; inflation
forecasts, which are more controllable than the inflation rate itself, hence become an implicit
intermediate target, and as such also shift into the focus of public scrutiny.
In the euro area, the Treaty assigns to monetary policy the primary objective of maintainingprice stability. While the European Central Bank (ECB) did not adopt an inflation targeting
framework as such, its Governing Council nonetheless provided a quantification of price
stability, defined as an annual increase in the HICP below, but close to 2% over the medium
term for the euro area as a whole (ECB, 2003). It hence set a reference value, serving
concurrently as a yardstick for the assessment of the ECB’s past performance, and as a point
of orientation for inflation in the future. Diron and Mojon (2005) compare the forecast
performance of inflation targets of the euro area and seven inflation targeting countries, to
model-based and published inflation forecasts respectively. They find that the central banks’
targets generally match or outperform the alternatives considered, and conclude that there are
substantial benefits in adopting the stated target as a forecast, in particular over longer
horizons.
A number of authors have shown that simple inflation forecasting models have often
performed well with respect to more elaborated ones. Fritzer et al. (2002) find that for
Austrian inflation, univariate (ARIMA) forecast models outperform multivariate (VAR)
models at short horizons; the opposite is however true at more distant horizons (8 to 12
months). Considering a set of univariate and multivariate euro area inflation forecasting
models, Hubrich (2005) testifies the good overall performance of the autoregressive model,
which even turns out to be the most accurate model at a 12-month forecast horizon. For the
US, Stock and Watson (1999) also note the “surprisingly” strong performance of
autoregressions and random walk models. Also for the US, Atkeson and Ohanian (2001)
evaluate the accuracy of Phillips curve-based inflation forecasting models based on the non-accelerating inflation rate of unemployment. They find that none of three sets of inflation
forecasts considered (forecasts published by the Federal Reserve Board, as well as two model-
based forecasts), succeeds in systematically outperforming the random walk model used by
them as benchmark.
While most economic agents’ and observers’ ultimate interest in inflation centres on the
headline rate, the question whether it should be forecasted directly (‘top-down’) or indirectly
(‘bottom-up’), has received a lot of attention among forecasters.4
Hubrich (2005) reviews
theoretical considerations behind the two approaches, noting that the “results from asymptotic
theory and small sample simulations do not give a clear answer regarding [their] relative
forecast accuracy” (p.122). In the absence of clear theoretical guidance, the judgement thusessentially becomes an empirical one. A number of authors have empirically assessed the
relative forecast performance of direct and indirect approaches to forecasting. Overall, their
findings do not allow systematically favouring one method over the other, let alone discarding
one. Indeed, the findings seem to be sensitive, inter alia, to the geographical entity, the types
of models considered, the forecast horizon, the data frequency and the transformation of the
price index series, as well as the periods chosen for model selection and evaluation. Fritzer et
al. (2002) construct two series for Austrian headline inflation by aggregating component-
specific inflation forecasts generated by ARIMA and by VAR models, respectively. In the
ARIMA case, the indirect approach generally beats the corresponding direct ARIMA
4 Another indirect way of forecasting, not considered here, would be to forecast individual Member States’ series
and to aggregate them to obtain a euro area forecast series (see Benalal et al. (2004)).
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 7/37
5
forecasts. In the VAR case, however, the indirect approach becomes only slightly superior at
distant horizons. Moser et al. (2004) find that indirect forecasts for Austrian inflation appear
to be somewhat more precise than the best direct inflation forecast considered; their model
competition is based on ARIMA, VAR and factor models. Den Reijer and Vlaar (2006),
considering AR, VAR and VECM models, find that for the Netherlands the indirect approach
is only superior at closer forecasting horizons (up to 7 months). For the euro area, they findthat the indirect approach outperforms the direct approach on all forecast horizons evaluated
(18 months). Meanwhile, Benalal et al. (2004), considering both univariate and multivariate
(VAR, Bayesian VAR and single equation) models, find that for the euro area (as well as for
Germany, France, Italy and Spain) the direct HICP forecasts improve upon indirect forecasts
for horizons beyond 12 months. For closer horizons their results are however less evident.
Hubrich (2005) finds that the indirect approach primes for forecasts of euro area inflation one-
month ahead, but that it does not generally improve upon direct forecasts at forecast horizons
of 6- or 12-months, which are more relevant from a policy point of view. In a more recent
study, Bermingham and D’Agostino (2010) find that aggregating forecasts of inflation
components (at different levels of disaggregation) yields more accurate forecast than direct
forecasting for both the euro area and US inflation. However, they stress the necessity of specifying an appropriate model for each individual component first.
3. Euro area inflation between 1996 and 2010
3.1 ANNUAL INFLATION PATTERNS
This section provides a brief description of euro area (16 Member States) inflation
developments between 1996:01 and 2010:12.
5
Figure A.1 in the annex displays the price levelindex (in logarithmic transformation), as well as monthly and annual changes respectively for
all series. Annual headline inflation, measured by year-on-year changes in the overall
harmonised index of consumer prices (HICP), has averaged around 1.9% over the sample
period. It steadily declined from 1997 until the introduction of the euro in 1999. It then
rebounded and remained broadly stable between the early 2000's and mid-2007, hovering in a
range of 1.6% to 3.1%. In the following period and until the end of the sample, inflation was
characterised by large swings: it climbed to 4.0% in July 2008, plunged to -0.6% one year
later and progressively returned to around 2% by the end of the sample period. These swings
were primarily the result of the patterns of global commodity prices, which impacted directly
on food and energy inflation in the euro area during that period, and coincided with the "great
recession" period triggered by the financial crisis that started in mid-2007 and intensified afterthe failure of Lehman Brothers in September 2008.
A notable feature of annual euro area inflation developments is that price indices in the
different categories have on average grown at a very different speed in the period considered
(see table 1). The non-energy industrial goods component, which on average has accounted
for almost one third of the consumer basket, is the only category for which prices have grown
at an average pace below 2% a year (0.8%). Low inflation in this category most likely reflects
the strong external competitive pressures that exist for many manufactured products, as well
as quality improvements in the product items covered. At the other extreme, with an average
of 4.0% a year, energy prices have grown fastest over the period considered, followed by
5The HICP indices are published by Eurostat.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 8/37
6
processed food (2.4%) and services (2.2%). Energy inflation has also been the most volatile,
with year-on-year rates falling into a 31 percentage point-range, from a high of 17.0% (July
2008) to a low of -14.2% (July 2009). Unprocessed, and to a lesser extent processed food
inflation, also oscillated considerably. Annual inflation of services and non-energy industrial
goods respectively has evolved in a smoother way, thus lessening the volatility of overall
inflation. Services and processed food are the only components which never experiencednegative annual price growth, having fallen to respective minima of 1.2% and 0.4%. It is also
confirmed that inflation in the core categories – non-energy industrial goods, services and to a
lesser extent processed food – generally fluctuates less and within narrower bands. Finally, it
is noteworthy that for most components the extreme values in annual inflation occurred after
2007.
Tab le 1: Desc rip tive sta tistics of euro a rea inflat ion (1996:01-2010:12)6
cp00 enrgy foodproc foodunp igoodsxe serv
Mean 1.93 4.03 2.37 2.14 0.76 2.23
Median 2.01 3.33 1.96 1.98 0.76 2.35
Maximum 4.06 17.01 7.20 8.93 1.78 3.35
Minimum -0.65 -14.25 0.35 -1.65 -0.12 1.18
Range 4.70 31.26 6.85 10.57 1.90 2.17
Std Dev 0.77 6.44 1.46 2.18 0.35 0.52
cp00 enrgy foodproc foodunp igoodsxe serv
Mean 0.16 0.34 0.19 0.18 0.08 0.18
Median 0.18 0.30 0.13 0.16 0.13 0.14
Maximum 1.11 4.10 1.18 3.47 2.48 1.14
Minimum -0.83 -4.90 -0.12 -1.59 -3.14 -0.75
Range 1.94 8.99 1.31 5.06 5.62 1.89
Std Dev 0.29 1.44 0.22 0.73 0.89 0.37
Annual inflation (%)
Monthly inflation (%)
Source: own calculations based on Eurostat
3.2 INTRA-YEAR VARIATION OF INFLATION
Some of the inflation series, in particular non-energy industrial goods and services, display
clear seasonal patterns, which shape the profile of monthly headline inflation in proportion to
their weight in the HICP. The additional presence of seasonal trends in some of the series also
generates a seasonal pattern in annual inflation.
Figure 1 below (left panel) shows the average monthly profile for overall inflation over the
sample period (applying the 2010 weights). Monthly headline HICP inflation is positive in
most months (+0.16% on average). Notable exceptions are the sales months January and July,
when prices have declined on average by 0.32% and 0.14% percent, respectively. Both drops
in the price level offset increases of a similar magnitude in the preceding months, i.e.
December (+0.31%) and June (+0.10%), respectively. About two thirds of the annual price
increases occur in the first half of the year, with March (+0.54%), April (+0.34%) and
February (+0.31%) typically being the months with the strongest price increases.
6Note that one observation is lost compute monthly inflation and 12 are lost to compute annual inflation.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 9/37
7
Figure 1: Breakdown of intra-year inflation profile and contributions to annual inflation
Typical inflation Profile(Based on 1996:01-2010:12 averages and 2010 weights)
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Energy Foodunp Foodproc NEIG Services HICP
Euro Area headline inflation and contribution(Based on 1997:01-2010:12)
-2
-1
0
1
2
3
4
5
Jan-97 Jan-99 Jan-01 Jan-03 Jan-05 Jan-07 Jan-09
Non-ene rgy ind . goods E ne rgy
Services Unprocessed food
Processed food Headline HICP Source: own calculations based on Eurostat
The seasonal plots of monthly inflation for the five components (figure A.2 in the annex)
provide additional insights regarding intra-year variation. A seasonal profile is only hardly
discernible for energy inflation. Processed food prices typically make a jump in January,
which is likely to reflect the scarcity (or unavailability) at that time of the year of certain itemsincluded in the index. However, intra-year variation is very low otherwise. In contrast,
unprocessed food displays a very pronounced seasonal pattern. Similar to processed food,
prices for unprocessed food make a distinctive leap in January, but decline from June to
August. The high volatility of monthly inflation can also result from disruptions such as
weather-related uncertainties of crop production or animal and vegetal diseases. Non-energy
industrial goods prices exhibit the most pronounced seasonal profile, with significant price
drops in the sales months January and July. Two interesting developments are worth pointing
out: first, monthly price decreases in January and July have become larger over time, while,
secondly, the price increases which have followed in March, and in September and October,
respectively, have accelerated over the years. This suggests that summer and winter sales may
have become more aggressive over time, with firms offering increasingly large discounts,while the following rebound in prices has also become stronger. Services inflation also
displays a quite marked intra-year variation, inter alia determined by holiday periods.
4. Model selection strategy
This section sets out the detailed strategies that are followed to select the models, which will
ultimately be evaluated on the basis of their forecast precision. The first sub-section (4.1)
introduces the benchmark models; the ARMA and ADL-model selection algorithms are
presented in sub-sections 4.2 and 4.3, respectively. All model selection procedures are appliedto the headline inflation series for the specification of direct inflation forecasting models.
Separately, they are applied to each of the five inflation component series to specify the
component-specific models which are the basis of the indirect inflation forecasts.
4.1 RULES OF THUMB AND ECONOMETRIC BENCHMARK MODELS
To obtain a yardstick for judging the forecasting accuracy of candidate models, a group of six
benchmark models is considered. It contains four econometric models, including three
autoregressive models - AR (1), ARMA (1,1) and AR (12) – and a seasonal average model,
setting monthly inflation in a particular month equal to the average inflation observed in thatmonth in the past. For headline inflation, two additional "rules of thumb" are considered: the
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 10/37
8
first is the ECB’s definition of price stability, discussed in the survey. For the purpose of this
paper, and following Diron and Mojon (2005), the ECB’s inflation objective is defined as an
annual inflation rate of 1.9% (denoted “ECB-target” in the following). Since the objective of
price stability applies to the medium-term, the target should be expected to improve relative
to other models as the length of the forecast horizon increases. Target misses at close forecast
horizons should accordingly not be ascribed to central bank decisions, as monetary policy hasno corrective impact on inflation in the very short-run. The second rule of thumb ("naïve"
model) simply corresponds to the year-on-year inflation rate prevailing at the moment the
forecast is made; it hence does not take account of temporary factors that may explain the
current inflation rate. Nevertheless, the “model” is easy to implement, as it only requires
knowledge of current inflation, and also recognises that the last available observation
generally constitutes a natural starting point and orientation for any forecast. The naïve
approach could even be expected to perform reasonably well at short horizons when there is
inflation persistence or at more distant ones when the last available inflation rate is already
close to target.
4.2 ARMA-MODEL SELECTION ALGORITHM
Autoregressive forecasting models rely on identifying the data-generating process (DGP)
underlying the inflation series. This “essentially agnostic” (Meyler et al. (1998)) way of
forecasting is purely mechanical rather than fundamentals-based, and can therefore in some
ways be criticised as simplistic. Nonetheless, univariate models have often been found to
perform well in short-run inflation-forecasting (see survey), and have the practical advantage
that data requirements are limited to the series of interest.
The atheoretical nature of univariate models leaves many liberties when specifying a forecast
model. This, however, risks making the selection somewhat arbitrary and non-transparent.Applying a Box-Jenkins model-selection approach could indeed lead to find models with a
higher (in-sample and out-of-sample) fit, but would also introduce a large scope of discretion,
resulting in different models depending on the individual forecaster (see Meyler et al. (1998)
for a discussion). For that reason, it is proposed to pick the models retained for the evaluation
stage through an automated selection procedure based on objective selection (information)
criteria (cf. Meyler et al. (1998), and Hubrich (2005)). Information criteria essentially differ in
the severity by which they sanction the inclusion of additional parameters in a model, relative
to its fit. Of the two mainly used criteria – Schwarz (SIC) and Akaike (AIC) – the SIC
generally leads to more parsimonious lag orders, while the AIC tends to overfit a model by
putting a lower penalty on additional parameters than the SIC (for a discussion see Lütkepohl
and Krätzig (2004)). In this paper, a large set of autoregressive models are estimated. Thosewith the lowest AIC and the lowest SIC, respectively, are retained and carried forward to the
out-of-sample simulation exercise.
The order of an autoregressive integrated moving average (ARIMA) process is expressed as
(p, d, q)7, where p refers to the number of lags of the dependent variable and q to the number
of moving average (MA) terms, and d is the order of integration of the original series. While p and q will be determined by the minimum value of the information criterion, the differencing
decision is made on the basis of a separate stationarity analysis. The latter, presented in annex
2, has suggested using either the first-differenced series (i.e. monthly inflation)8
or seasonal
7 Seasonal lags and MA terms are not considered here.8 The unit root hypothesis is rejected for all first-differenced series by the Philips-Perron test. The ADF-test leads
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 11/37
9
differences of first-differenced series, for all series. For the present exercise, the first-
differenced log series (monthly inflation) is used, i.e. d=1; this is also in line with other
authors (e.g. Hubrich, 2005). Since monthly inflation now is the series of interest, the notation
ARMA (p, q) will be used in the following.
Four ARMA-specifications are considered. The first sample observation in first-differencedform is 1996:02 and 12 pre-sample observations have to be set aside to allow for a maximum
of 12 lags, while ensuring an equal number of observations across all estimations. The models
are hence estimated over the samples 1997:02-2010:12 (including the crisis) and 1997:02-
2007:12 (pre-crisis), respectively. The basic ARMA specification is given by
(U.1) t t L Bc L A e m )()( += ,
where p
p L L L A a a ---= L11)( and q
q L L L B b b ---= L11)( , with 11max= p and
11max=q , and µ t is the first difference of the log price index series under consideration (i.e.
monthly inflation). L is the lag-operator. The selection procedure considers only full sets of adjacent lags (i.e. including all lags up to the maximum). Determining p and q requires the
estimation of 144 ((p+1)*(q+1)) models for a given selection criterion. Note that pure AR
(i.e. q=0) and pure MA (i.e. p=0) models are special cases that are considered under this
selection procedure. The second specification, which is obtained by augmenting specification
(U.1) by a full set of seasonal dummies (SD)9, is likely to be particularly relevant for series
exhibiting a distinct intra-year pattern. It is expressed as:
(U.2) t d d t L BSDc L A e d m )()(
11
1++= å
As inflation in a particular month may be affected by monthly inflation one year earlier and
may be driven by seasonal regularities, specification (U.1) is augmented by the twelfth lag of
monthly inflation, becoming
(U.3)t t L Bc L A e m )()(12
+= ,
where 12
121
12 1)( L L L L A p
p a a a ----= L , with 11max= p , and B(L) is defined as above.
The twelfth lag of the dependent variable is thus explicitly introduced in the regressions,
while the choice of p and q is again left to the information criterion. This ARMA "with a
hole" is a special case of a (12, q) process, possibly with some intermediate lag coefficients
set to zero. A pure AR (12) process is thus also part of the specifications considered. In
analogy with specification (U.1), (U.3) is augmented by seasonal dummies, yielding thefollowing specification:
(U.4) t d d t L BSDc L A e d m )()(11
1
12 ++= å
The order of the processes eventually retained by the algorithm is shown in the tables A.6-
A.12 in annex 5.
to more ambiguous results.9 SD is equal to 1 in month d and equal to 0 otherwise. As the equation includes a constant, 11 seasonal dummies
are included.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 12/37
10
4.3 AUTO-REGRESSIVE DISTRIBUTED LAG MODELS
The third group of models exploits structural relationships observed in historical data between
inflation and different exogenous variables. These models can be broadly categorised,
according the type of exogenous variable included, as cost-push (supply-side) or demand-side
models.10
The former include cost-indicators as potential drivers of inflation and dependmainly on vertical production relations (e.g. energy prices strongly depend on oil prices). The
latter are essentially Philips-curve-type relationships, which explain inflation by (deviations
in) measures of actual or potential economic activity (e.g. output gap, unemployment,
industrial production): low demand leads to excess supply and, as firms need to liquidate their
products, to lower pricing power and eventually lower prices – and vice versa. Finally, other
variables, such as household surveys, may also exhibit leading indicator properties for
inflation.
4.3.1 INVESTIGATION OF "RELEVANCE" OF EXOGENOUS VARIABLES
Following Fritzer et al. (2002), an exploratory analysis is performed first, aiming to identifythe most relevant exogenous variables among a set of candidates. The selected variables will
enter the more refined ADL model-selection process described in the next sub-section.
A total of around 40 exogenous variables are considered here that are of monthly frequency
(i.e. no interpolated quarterly data are used), and can be classified in the following broad
categories, which can all be expected to drive inflation and/or act as leading indicators:
commodity prices; exchange rate; industrial production; industrial orders; surveys; labour
market . The individual series are listed in table A.4 (annex 3).
The following “pre-fitting” equations are estimated for the sample period 1997:02 to 2010:12:
(B.0)t
j
jk t i
i
j
it i
j
t xc e b m f m +++= åå=
--
=
--
12
0
12
0
1 ,
where the j superscript identifies the inflation series and x is the exogenous variable of
interest, which in general enters the equation as first-difference of the log-series (for some
series the percentage or absolute change is retained). Each equation is estimated for three
different displacements (k=1, 6, 12) of the exogenous variable, to assess whether it could
constitute a leading indicator for inflation. A contemporaneous equation (with no
displacement, i.e. k=0) is also fitted, as a feedback effect from inflation to the exogenous
variables used here can be ruled out on conceptual grounds and as such should not give rise to
endogeneity concerns. The joint significance of the exogenous variables is tested applying an
F-test. Under the null hypothesis it assumes that the coefficients on the exogenous variables
(the betas) are jointly equal to zero, such that the equation (B.0) reduces to an AR(12) model.
Table A.5 in the annex 3 reports the F-statistic and the probability value, as well as the
goodness of fit of the unrestricted specification, measured by the adjusted coefficient of
determination. For headline inflation the pre-fitting equations suggest that commodity prices
10 Monetary models are another category of models (not considered in this paper; see for example Hofmann,
(2006)), which could be used for longer horizons and rest on the idea that money is neutral in the long run,
following the Friedman-Schwarz insight that ‘inflation is always and everywhere a monetary phenomenon’. Therelationship between monetary growth and inflation has also shaped the Eurosystem’s monetary policy strategy,
in which monetary analysis constitutes the second of two pillars (ECB, 2003).
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 13/37
11
and activity variables (industrial production, orders and employment) are a priori relevant
leading indicators and significantly improve the fit of the regressions, contrary to exchange
rates and consumer surveys (with the exception of expected price trends over the next 12
months).11
On the basis of this pre-fitting analysis, the significant exogenous variables, which offer thelargest explanatory potential, are carried forward to the actual ADL-model selection. Some
variables that are considered particularly important from either a macroeconomic (e.g.
commodity prices, exchange rates) or sectoral (e.g. agricultural raw materials in the case of
food inflation) perspective, are maintained regardless of this pre-fitting outcome.
4.3.2 ADL-MODEL SELECTION-ALGORITHM
This section describes the selection procedure for ADL-models. The estimated process in its
most general form is given by the following ADL-model:
(B.1) t
q
j
jt j
p
i
jit i
jt xc e b m f m +++= åå
=
-
=
--
00
1 ,
with pmax
=11 and qmax
=25, allowing for a lead time of two years; µ and x are defined as in the
pre-fitting equation (B.0) above, except that the lag orders p and q are now determined via the
penalty function. Adding a full set of seasonal dummies to specification (B.1) yields the
following equation:
(B.2) t
q
j
jt j
p
i
j
it id
j
t xSDc e b m f d m ++++= ååå=
-
=
--
00
1
11
1,
The next specification modifies (B.1) by adding the twelfth lag of the dependent variable into
the equation:
(B.3)t
q
j
jt j
j
t
p
i
j
it i
j
t xc e b m m f m ++++= åå=
--
=
--
0
12
0
1
Augmenting it by a set of seasonal dummies, yields the following equation:
(B.4)t
q
j
jt j
j
t
p
i
j
it id
j
t xSDc e b m m f d m +++++= ååå=
--
=
--
0
12
0
1
11
1
For each inflation series and for each respective exogenous variable, the model is estimated
for all combinations of p and q up to their respective maximum, yielding a total of 312
(12*26) models per specification, including a set of pure AR models (i.e. no exogenous
variable) and a set of pure distributed lag models (i.e. no lag dependent variable) as special
cases. The combination of p and q returning the lowest information criterion is retained for
the evaluation stage. The AIC is now used as evaluating criterion, as it imposes a lower
penalty on additional variables than the SIC, and thus reduces the probability of dropping the
exogenous variable.
11An analogous exercise is performed for the five inflation components (available upon request).
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 14/37
12
Following the strategy described in section 4.2 (ARMA) and 4.3 (ADL) above, carries the
risk that a good forecast model is already discarded at the selection stage. To avoid this, all
candidate models initially considered could in principle be directly evaluated on their
forecasting accuracy - which after all is the ultimate feature of interest - without undergoing
this pre-selection stage. However, there are two good reasons to maintain a pre-selection:
first, it ensures that the set of candidate models is narrowed down to a more manageable size.Secondly, it screens models according to their fit, ensuring that a meaningful statistical
relationship exists between the variables.
5. Evaluation of the forecast accuracy during thecrisis
The candidate models picked by the algorithms are evaluated by their out-of-sample accuracy
measured by the recursive Root Mean Squared Forecast Error (RMSFE) presented in annex 4.
Each forecast horizon is evaluated over 48 observations, i.e. 4 years. The first recursive model
is estimated over the sample 1997:02 to 2005:07. The end date of the sample is gradually
extended until 2009:06, leaving 18 remaining observations for evaluation purposes.12
Overall,
the evaluation period was relatively calm until the end of 2007, but became more agitated
afterwards, when global oil and food prices started to hike and the financial crisis started.
Headline inflation between 2005:08 and 2010:12 averaged 1.94%, i.e. close to target, but was
very volatile (see table A.2.1 in annex 1 for descriptive statistics of the evaluation period).
5.1 DIRECT FORECASTS OF HEADLINE INFLATION DURING THE CRISIS
Table 2 (left panel) below summarises the accuracy of the main model groups, displayingtheir average RMSFEs over all 18 forecast horizons, as well as the average RMSFEs for three
blocks of six-month forecast horizons. Table A.6 in annex 5 displays the RMSFEs for each of
the 18 forecast horizons for all competing headline inflation models. The ADL-model
specification (i.e. B.1-B.4) included in the table, is the one yielding the lowest average
RMSFE.13
12One-step-ahead forecasts are thus evaluated over the period 2005:08 to 2009:07, while the 18-step-ahead
forecasts are assessed over the 48-month period between 2007:01 and 2010:12.13 When evaluating ADL models, a path for the exogenous variables needs to be specified. In this paper the out-
of-sample evaluations of the ADL models are based on conditioning assumptions that correspond to the true path
of the exogenous variable of interest. While this assumption of perfect foresight is of course very strong and
unlikely to be met in practice, it however allows focussing on the particular exogenous variables. With
assumption errors thus effectively being ruled out as a source of inflation forecast error, the RMSFEs obtainedshould be regarded as a lower bound and an indication of the margin of improvement for other models. The
results may hence help to focus on those variables that are particularly valuable as inflation predictors.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 15/37
13
Tab le 2: Averag e RMSFEs of ma in model c a tegories in crisis-sample (summary)
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks
ECB 1,12 1,19 1,19 1,17 (-) (-) (-) (-)
Naive 0,77 1,60 1,88 1,42 (-) (-) (-) (-)
AR(1) 0,71 1,17 1,24 1,04 0,71 1,17 1,24 1,04ARMA(1,1) 0,69 1,15 1,21 1,02 0,70 1,16 1,24 1,03
AR(12) 0,79 1,51 1,67 1,32 0,60 1,26 1,48 1,12
Seas. Dummies 0,57 1,09 1,23 0,96 0,57 1,09 1,23 0,97
ARMA (*)
Identical DGP 0,58 1,11 1,26 0,98 0,54 1,11 1,31 0,99
Component-specific DGP (-) (-) (-) (-) 0,48 1,02 1,20 0,90
ADL (oilusd) (*)
Identical model 0,35 0,75 0,95 0,68 0,33 0,70 0,89 0,64
Component-specific model (-) (-) (-) (-) 0,32 0,67 0,86 0,62
Combination (-) (-) (-) (-) 0,28 0,46 0,53 0,42
Direct forecast Indirect forecast
Note: (*) Refers to the model specification generating the lowest RMSFE of the respective model group
For headline inflation, the accuracy of forecasts produced by the ARMA-models selected by
an information criterion tends to be comparable to that of benchmark models. In particular,
the seasonal dummy-model appears difficult to beat, reflecting the importance of the intra-
year inflation pattern. The ECB-target improves in relative terms with the distance of the
forecast horizon, dominating all ARMA-models on horizons beyond twelve months, which
are more relevant from a monetary policy perspective. The naïve model is rapidly
outperformed by the ARMA-models, which is unsurprising in the light of the swings in
inflation observed during the evaluation period. Considering ADL-models reveals that theinclusion of commodity prices, in particular of oil prices, considerably reduces the forecast
errors over all horizons compared to ARMA and benchmark models. The RMSFE of the
model including dollar-denominated oil prices corresponds to around 70% of that of the
seasonal-dummy model and to around 60% of the static ECB-target forecast. Industrial
production, industrial orders and labour market data also improve the out-of-sample fit, albeit
to a lesser extent.
5.2 INDIRECT FORECASTS OF HEADLINE INFLATION DURING THE
CRISIS
Indirect headline inflation forecasts are obtained by aggregating forecasts of each of the five
components (generated by some component-specific models), weighted by their share in the
HICP in the relevant year. The aggregated inflation series obtained through this indirect
approach are then also evaluated by the RMSFE. While a large number of combinations of
component-forecasts are possible, six sets of aggregation are considered here: i) aggregations
of forecasts produced by the four econometric benchmark models; ii) an aggregation of
forecasts obtained by imposing the data-generation process of the most accurate direct
ARMA-model on all series; iii) an aggregation of ARMA forecasts, allowing for a
component-specific data-generation processes; iv) an aggregation of ADL-forecasts using
dollar-denominated oil prices as an explanatory variable for all components, imposing the
same model on all series; v) an aggregation of ADL-forecasts based on dollar-denominatedoil prices, now allowing for component-specific specifications; vi) an aggregation of those
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 16/37
14
forecasts, which have given rise to the lowest average RMSFE for each component, i.e.
allowing for component-specific models with different exogenous variables. Aggregations i),
ii) and iv) thus fit an identical model to each individual series, as to the headline series; the
other aggregations consider true component-specific models.
Before turning to the evaluation of the aggregated series, the results of the forecasting modelsfor the individual inflation components are briefly commented in this paragraph. The
candidate models for the respective components and their average RMSFEs over all 18
forecast period, and the average RMSFEs over three blocks of six-month horizons are
reported in tables A.8 to A.12 in the annex. The results show that the components attributed to
core inflation, which are less volatile and exhibit stronger regularities, also have the highest
out-of-sample fit. This is well-illustrated by the non-energy industrial goods and services
series on the one hand, which have the smallest forecast errors, and by the energy series on
the other, which has the largest. In the case of energy inflation, the highest forecast accuracy –
albeit low by the standards of the other components – is achieved by an ADL-model that
includes oil prices.14
For processed food inflation, the ADL-model including the euro-dollar
exchange rate produces the smallest forecast errors.15
For unprocessed food inflation, thehighest forecast accuracy is obtained by including activity variables and industrial order data,
which dominate commodity prices, such as agricultural raw materials and cereals. The strong
seasonal regularities observed for non-energy industrial goods inflation generally facilitate
the forecasting of this series, as is confirmed by the generally low RMSFEs relative to other
inflation series. Adding exogenous variables can in some cases (e.g. industrial production or
labour market variables) reduce the forecast errors even further, albeit only marginally. For
services inflation, the forecast accuracy of the simple seasonal-dummy model is similar to that
of the best ARMA-models picked by the algorithm. An ADL-model that includes the
unemployment rate slightly improves the out-of-sample fit with respect to benchmarks and
ARMA-models.
The RMSFEs obtained by the indirect approach are reported in table 2 (right panel) and in
more detail at the bottom of table A.6 in the annex. Except for the AR (12) model, indirect
forecasts based on benchmark models do not yield better forecasts than the corresponding
direct forecast. The indirect forecast obtained from applying the DGP of the best direct model
to all component-series, does not improve upon the direct forecast overall: while the forecast
accuracy is slightly higher up to six-months ahead, it deteriorates over horizons further than
12 months away. In contrast, when allowing for differentiated component-specific DGPs, the
indirect approach clearly outperforms the best direct ARMA-model. This supports the
conclusion of Bermingham and D’Agostino (2010), who stress the importance of specifying
an appropriate model for each component. Indirect forecast resulting from ADL-models
including oil prices dominate the corresponding direct approach, regardless of whether anidentical model is imposed or component-specific models are fitted. The final aggregation,
which is effectively a multivariate forecasting set-up, relies on dollar-denominated oil prices
(energy), the exchange rate (processed food), industrial production of consumer goods
(unprocessed food and non-energy industrial goods) and the unemployment rate (services). It
generates a lower average RMSFE than all direct and indirect approaches considered: the
average RMSFE corresponds to 36% of that of the ECB-target and 30% of that of the naïve
14Interestingly, dollar-denominated oil prices constitute a better predictor for energy inflation in the euro area
than euro-denominated oil prices (the same holds for headline inflation).15
It is also worth noting that the agricultural raw materials price index is not retained by the ADL algorithm.While the unprocessed food price index is retained by the algorithm, the resulting model does not yield better
forecasts.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 17/37
15
model. Relative to the best direct forecast (ADL-model including dollar-denominated oil
prices), the average RMSFE corresponds to 63%. This aggregation also improves in relative
terms with the length of the forecast horizon: while for the six months ahead, the average
RMSFE of the indirect approach represents around 80% of that of the best direct model, it
drops to around 56% on horizons that are between 13 and 18 months ahead.
6. Evaluation of the forecast accuracy before theeconomic and financial crisis
The evaluation period in section 5 has to a large extent coincided with the financial and
economic crisis. As the latter is likely to have impacted the conclusions, the validity and
robustness of the latter for other periods may thus a priori be questioned. Indeed, the largeswings in global oil and food prices after 2007 and the disruptions in economic activity
caused by the crisis itself had a major impact on the euro area inflation profile, as illustrated
in section 3, but also on private and public institutions’ forecasts, which had to be
significantly revised (cf. table A.1). To assess the robustness of the results over time, this
section considers how the forecasting performance of the models considered above would
have been different – in both absolute and relative terms – if the selection and evaluation
process had been carried out before the start of the crisis.
To assess the pre-crisis forecasting accuracy, the last three years of the sample are dropped,
and the model selection algorithms described in section 4 are run over the sample 1997:02 to
2007:12. Maintaining an evaluation period of 48 months, the first recursive sample ends in
2002:07, while the last sample runs until 2006:06.16
While average headline inflation between
2002:08 and 2007:12 was on average somewhat higher than during the evaluation period
considered in section 5, namely at 2.17%, the volatility, measured by the range or the standard
deviation, was considerably lower (see table A.2.2 in the annex). On average, inflation in all
categories with the exception of energy was higher in the pre-crisis evaluation period.
Meanwhile, inflation volatility was lower for all components, except for unprocessed food
and non-energy industrial goods inflation.
6.1 DIRECT FORECASTS OF HEADLINE INFLATION PRE-CRISIS
Table 3 (left panel) summarises the performance of the main model groups, showing their
average RMSFEs for the entire evaluation period and for three six-month periods. Table A.7
in the annex displays the pre-crisis RMSFEs of the benchmarks and all selected directheadline inflation models. The ADL-models reported in the table correspond to the
specification (B.1-B.4) that yielded the lowest average RMSFE. The precise specifications of
the models evaluated before the crisis are consequently not necessarily the same as the ones
evaluated during the crisis.17
16One-month ahead forecasts are assessed over the period 2002:08 to 2006:07, while the 18-month forecast
horizon is evaluated over the period 2004:01 to 2007:12.17
It is noteworthy that some exogenous variables may be retained by the algorithm in one period, but not in theother. In the case of processed food inflation for example, agricultural raw materials prices would have been
included in an ADL-model prior to the crisis, but not thereafter.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 18/37
16
Tab le 3: Ave rag e RMSFEs of m a in model c a tegories in pre-c risis-samp le (summary)
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks
ECB 0,34 0,32 0,34 0,33 (-) (-) (-) (-)
Naive 0,29 0,36 0,36 0,34 (-) (-) (-) (-)
AR(1) 0,44 0,52 0,39 0,45 0,47 0,55 0,38 0,47 ARMA(1,1) 0,42 0,49 0,38 0,43 0,47 0,55 0,32 0,45
AR(12) 0,31 0,35 0,35 0,34 0,29 0,35 0,33 0,32
Seas. Dummies 0,30 0,40 0,42 0,37 0,31 0,40 0,40 0,37
ARMA (*)
Identical DGP 0,28 0,35 0,37 0,33 0,30 0,48 0,55 0,44
Component-specific DGP (-) (-) (-) (-) 0,28 0,36 0,34 0,33
ADL (oilusd) (*)
Identical model 0,22 0,26 0,23 0,24 0,22 0,30 0,31 0,28
Component-specific model (-) (-) (-) (-) 0,21 0,26 0,24 0,24
Combination (-) (-) (-) (-) 0,20 0,23 0,25 0,23
Direct forecast Indirect forecast
Note: (*) Refers to the model specification generating the lowest RMSFE of the respective model group
Comparing the forecast accuracy of the different models highlights the relatively strong
performance of the ECB-target. On average it dominates all benchmark and ARMA-models,
essentially due to its good performance over horizons beyond 8 months. It is also interesting
to note that, contrary to other models, the forecast errors of the ECB-target remain stable and
even decline with the length of the f orecast horizon, consistent with the medium-term
character of the price stability objective.18
Finally, the ECB-target also performs reasonably
well compared to ADL-models, in particular at more distant forecasting horizons. The
average RMSFE of the best direct model, an ADL-model including oil prices, corresponds to72% of the ECB-targets’ RMSFE. Overall, this would have supported the conclusion of Diron
and Mojon (2005) that the ECB-target constitutes a good rule of thumb over medium-term
horizons. In general, the performance of the other benchmark models was also relatively
strong compared to ARMA and ADL-models selected on the basis of the penalty function.
The naive model, for example, performed comparatively well at horizons below 12 months,
owing to the stability of inflation during that period. These results corroborate the conclusions
of other studies (cf. survey) stressing the good performance of simple inflation forecasting
models.
6.2 INDIRECT FORECASTS OF HEADLINE INFLATION PRE-CRISIS
Indirect forecasts for the pre-crisis period are obtained from the same six aggregation sets as
in section 5.2; their RMSFEs are displayed in table 3 (right panel) and in more detail at the
bottom of table A.7 in the annex.19
Again, the indirect forecasts produced by the benchmark
models cannot outperform their corresponding direct equivalent. Also in line with the results
of the crisis period, the ARMA-based indirect approach leads to lower forecast errors than the
direct equivalent, on the condition that component-specific DGPs are allowed. In contrast to
the crisis period, indirect forecasts based on ADL-models including oil prices cannot
18 The rise in the RMSFE for forecasts beyond 16 months can essentially be attributed to the strong pick-up in
inflation at the end of the sample in 2007, which does not impact on the evaluation of closer forecast horizons.19 The evaluations of the forecasting models for the individual inflation components are reported in tables A.8 to
A.12 in the annex.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 19/37
17
outperform their direct equivalent. Imposing an identical model on all components actually
increases the size of the forecast errors with respect to the direct approach, while the
aggregation of forecasts of component-specific models produces comparable results to those
of the direct model.
The indirect inflation forecast obtained from the best component-specific models, allowingfor the inclusion of exogenous variables,
20yields an average RMSFE that is only marginally
below that of the best direct model. The better performance of the indirect inflation forecast is
essentially due to the better performance for horizons up to 12 months ahead. It should also be
considered in the light of the strong underlying assumptions: in particular, it requires the
knowledge of the path of three exogenous variables, while the best direct approach “only”
requires the knowledge of oil prices. The aggregation of the best component-specific forecasts
outperforms the rules of thumb (ECB-target and naïve), but by a narrower margin than during
the crisis: the average RMSFE is about 70% of that of the numerical benchmarks.
6.3
HOW DID THE CRISIS AFFECT THE ABSOLUTE AND RELATIVEFORECASTING ACCURACY?
Table 4 reports the ratio of the pre-crisis average RMSFE to the crisis average RMSFE for
comparable models. A value below 100% indicates that forecast errors of the given model
group were lower before the start of the crisis. Comparing the forecast accuracy of the
different groups of direct and indirect headline inflation models in both evaluation periods
reveals that all of them have deteriorated after the outbreak of the crisis, confirming the
impression that inflation forecasting has in general become more difficult in that period. This
is particularly well-illustrated by the rules of thumb and the benchmark models whose
precision significantly worsened: for example, the average RMSFE of the naïve model rose
from 0.34 before the crisis to 1.42, implying that taking the current inflation rate as aninflation forecast yielded four times larger errors with the crisis. The ECB-target forecast also
worsened significantly with the crisis, as the average RMSFE increased from 0.33 to 1.17.
20
Dollar-denominated oil prices (energy and non-energy industrial goods); exchange rate (processed food); euro-denominated oil prices (unprocessed food); services inflation is best modelled by an ARMA models including
seasonal dummies.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 20/37
18
Tab le 4: Pre-c risis ave rage RMSFEs rela tive to c risis ave rage RMSFEs (in %)
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks
ECB 29,9 26,7 28,4 28,3 (-) (-) (-) (-)
Naive 37,5 22,4 19,3 23,8 (-) (-) (-) (-)
AR(1) 62,6 44,1 31,8 43,4 66,3 46,9 30,8 44,9ARMA(1,1) 61,1 42,3 31,1 42,1 67,6 47,3 25,7 43,2
AR(12) 39,1 23,0 21,0 25,4 47,8 27,9 22,4 29,0
Seas. Dummies 52,7 36,3 34,3 38,7 54,6 36,5 32,8 38,5
ARMA (*)
Identical DGP 48,3 31,2 28,9 33,6 55,0 43,0 42,2 44,8
Component-specific DGP (-) (-) (-) (-) 58,2 34,8 28,1 36,0
ADL (oilusd) (*)
Identical model 61,3 34,6 24,8 34,7 66,7 42,9 34,8 43,3
Component-specific model (-) (-) (-) (-) 67,8 38,2 27,2 38,1
Combination (-) (-) (-) (-) 72,2 49,1 46,6 53,1
Direct forecast Indirect forecast
Note: (*) Refers to the model specification generating the lowest RMSFE of the respective model group
Compared to the benchmark models the weakening of more sophisticated models (i.e. ADL
and aggregations) has been relatively contained; as a consequence the rewards from investing
in them, rather than relying on simple models have increased. Put the other way round, the
attractiveness of resorting to simple rules of thumb has fallen with the crisis: evaluated during
the crisis, the average RMSFE of the dollar-denominated oil-price ADL-model (based on the
assumption of perfect foresight of the path of the exogenous variable), accounted for 59% of
the RMFSE of the ECB-target model. Before the crisis, the corresponding value was 72%.
Relative to the average RMSFE of the naïve model, the RMSFE of the oil-price ADL-modeldeclined from 70% before the crisis to 48% during the crisis. Note that the relative worsening
of benchmarks is not a consequence of the average inflation level during the two respective
evaluation periods (it was further away from the ECB-target in the pre-crisis period), but of
the more irregular inflation profile.
A look at the forecasts for the individual inflation components confirms that the deterioration
of forecasting models was broad-based, and not limited to one particular component (see
tables A.8-A.12). While in the case of energy inflation, the accuracy of the ADL-model
including dollar-denominated oil prices only deteriorated marginally with the crisis, for
processed food , all model groups produced substantially lower average RMSFEs before the
crisis. This is not surprising, given that this category was particularly affected by the globalfood price shock starting in 2007. For unprocessed food , the results are mixed: ADL-models
including oil and energy variables performed best before the crisis, while models that include
activity variables produced more accurate forecasts during the crisis. Services inflation
forecasting models have also deteriorated, regardless of the model category. For non-energy
industrial goods inflation, ARMA-models actually generated more accurate forecasts during
the crisis.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 21/37
19
7. Conclusion
This paper has analysed how the financial and economic crisis has affected inflation
forecasting in the euro area. Specifically, three groups of inflation forecasting models (rules
of thumb and benchmark models; autoregressive moving average models; autoregressive
distributed lag models) were evaluated under two distinct forecasting approaches: under the
first one, the models were applied directly to headline inflation, while under the second one,
component-specific forecasts were generated first and subsequently aggregated to obtain an
indirect forecast of headline inflation.
The paper’s first objective has been to compare the accuracy of different inflation forecasting
models. Regarding the direct ones, it finds that ARMA-models determined via a penalty
function generally perform at least as well as econometric benchmark models or rules of
thumb. Adding exogenous explanatory data in an ADL-setting in many cases improves the
forecast precision further. The best direct forecasts are obtained when including oil prices,
which play a key role in explaining and predicting inflation. The gains obtained from relying
on ADL-models are particularly large during the crisis period. As far as the indirect approach
is concerned, the forecasts are generally at least as accurate as those generated by the direct
method. However, the magnitude of the gains depends much on what precisely is being
aggregated: indeed, it is only once appropriate specifications for the component-specific
ARMA or ADL-models have been found, that the indirect forecasts dominate the
corresponding direct model. Indirect forecasts perform best when the information set is
widened, allowing different exogenous explanatory variables to enter in each component-
specific model. It should however be noted that in the pre-crisis period the best indirect
forecasts only marginally beat the best direct forecasts, and that this can partly be related to
the strong conditioning assumptions. Yet, even if a systematic superiority of the indirect
forecasting approach would be difficult to assert on the mere basis of the results of this paper,its use in practice nonetheless seems fully warranted and justifies further work on the
refinement of component-specific models.
An important lesson from the literature has been that the forecasting accuracy of models is
sensitive to the period in which they are estimated and evaluated, and that conclusions
regarding their merit may consequently be short-lived. The second objective of the paper has
been to illustrate this time-sensitivity and to gauge the impact of the crisis by repeating the
comparative assessment over two different sample periods. The rise in forecast errors of all
models, confirms that inflation forecasting has become substantially more difficult after the
onset of the crisis. However, the deterioration of the different models has been uneven: in
general, the performance of slightly more elaborated models – i.e. ARMA- and ADL-modelsselected on the basis of a penalty function – and in particular of indirect forecasts, has
improved during the crisis relative to simple econometric benchmarks or rules of thumb.
Indeed, while the model comparison during the pre-crisis period would have backed the
conclusions of earlier studies emphasising the virtues of simple models – including the central
bank’s inflation target – the comparative evaluation during the crisis would cast doubts on
those very same conclusions.
Future analysis on the specification of inflation forecasting models will need to take into
account the distortions caused by the commodity price swings and by the crisis after 2007. If
the choice of the forecasting model is made on the basis of past observations, e.g. by
performing an out-of-sample evaluation as in this paper, the question will arise how inflation
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 22/37
20
data from the crisis should be included in the model estimation, but also whether it seems
right to use the crisis as an evaluation period.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 23/37
21
References
Atkeson, Andrew and Lee E. Ohanian, 2001, Are Phillips Curves Useful for Forecasting
Inflation?, Federal Reserve Bank of Minneapolis Quarterly Review, Vol. 25, No.1, Winter
2001, pp. 2-11.
Benalal, Nicholai, Juan Luis Diaz del Hoyo, Bettina Landau, Moreno Roma and Frauke
Skudelny, 2004, To aggregate or not to aggregate? Euro area inflation forecasting, ECB
Working Paper 374, European Central Bank.
Bermingham, Colin and Antonello D’Agostino, 2010, Understanding and Forecasting
Aggregate and Disaggregate Price Dynamics, Research Technical Paper 8/RT/10, Central
Bank of Ireland, August 2010 (February 2011 update).
Bernanke, Ben and Frederic Mishkin, 1997, Inflation Targeting: A New Framework for
Monetary Policy, Journal of Economic Perspectives – Volume 11, Number 2, Spring 1997,
97-116.
Den Reijer, Ard and Peter Vlaar, 2006, Forecasting inflation: an art as well as a science, de
Economist, 154, 19-40.
Diron, Marie and Benoît Mojon, 2005, Forecasting the central bank’s inflation objective is a
good rule of thumb, ECB Working Paper 564, European Central Bank.
ECB, 2003, The ECB's monetary policy strategy, ECB Press release, 8 May 2003
Fritzer, Friedrich, Gabriel Moser and Johann Scharler, 2002, Forecasting Austrian HICP and
its Components using VAR and ARIMA Models, OeNB Working Paper 73, July 2002.
Hammond, Gill, 2011, State of the art inflation targeting – 2011, Centre for Central Banking
Studies, Handbook – No. 29 (February 2011 version), Bank of England.
Hofmann, Boris, 2006, Do money indicators (still) predict euro area inflation?, Deutsche
Bundesbank Discussion Paper No. 18/2006.
Hubrich, Kirstin, 2005, Forecasting euro area inflation: Does aggregating forecast by HICP
component improve forecast accuracy?, International Journal of Forecasting, 21, 119-136.
Lütkepohl, Helmut and Markus Krätzig, 2004, Applied Time Series Econometrics, Cambridge
University Press.
Meyler, Aidan, Goeff Kenny and Terry Quinn, 1998, Forecasting Irish Inflation using
ARIMA models, Research Technical Paper 3/RT/98, Central Bank of Ireland, December 1998.
Moser, Gabriel, Fabio Rumler and Johann Scharler, 2004, Forecasting Austrian Inflation,
OeNB Working Paper 91, September 2004.
Stock, James and Mark Watson, 1999, Forecasting inflation, Journal of Monetary Economics
44 (1999) 293-335.
Svensson, Lars E. O., 1997, Inflation forecast targeting: Implementing and monitoringinflation targets, European Economic Review, Elsevier, vol. 41(6), pages 1111-1146, June.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 24/37
22
Annex 1: Infla tion forec asts and presenta tion of data
Tab le A.1: Infla tion forec asts for 2008 and 2009 by ECB/Eurosystem, Europ ean
Co mmission a nd Survey o f Professiona l Forecasters
2008 2009
December 2006 1,9
March 2007 2,0
June 2007 2,0
September 2007 2,0
December 2007 2,5 1,8
March 2008 2,9 2,1
June 2008 3,4 2,4
September 2008 3,5 2,6
December 2008 3,3 1,4March 2009 0,4
June 2009 0,3
September 2009 0,4
December 2009 0,3
European Commission Spring and Autumn forecasts
2008 2009
Autumn (Nov 2006) 1,9
Spring (May 2007) 1,9
Autumn (Nov 2007) 2,1 2,0
Spring (April 2008) 3,2 2,2Autumn (Oct 2008) 3,5 2,2
Spring (May 2009) 0,4
Autumn (Nov 2009) 0,3
Survey of Professional Forecasters
2008 2009
November 2006 1,9
May 2007 1,9
October 2007 2,0 2,0
May 2008 3,1 2,1
October 2008 3,4 2,2
May 2009 0,5
October 2009 0,3
Average inflation 3,3 0,3
ECB and Eurosystem staff macroeconomic projections for the euro area
(*)
Note: (*) The figure refers to the mid-point of the published forecast. June and December forecasts are prepared
by ECB and Eurosystem staff. March and September forecasts are propared by ECB staff.
Source: ECB Monthly Bulletin, various editions.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 25/37
23
Figure A.1. HICP indices, annual and monthly inflation; overall and by component
(1996:01-2010:12)
-2
-1
0
1
2
3
4
5
6
4.40
4.44
4.48
4.52
4.56
4.60
4.64
4.68
4.72
1996 1998 2000 2002 2004 2006 2008 2010
MC P00 Y CP00 C P00_LOG
cp00
p e r c e n t
l o g of i n d ex
-16
-12
-8
-4
0
4
8
12
16
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
5.0
1996 1998 2000 2002 2004 2006 2008 2010
MEN RGY Y ENRGY EN RGY _LOG
enrgy
p e r c e n t
l o g of i n d ex
-2
0
2
4
6
8
10
12
4.40
4.45
4.50
4.55
4.60
4.65
4.70
4.75
1996 1998 2000 2002 2004 2006 2008 2010
MFOODUNPYFOODUNP
FOODUNP_LOG
foodunp
p e r c e n t
l o g of i n d ex
-1
0
1
2
3
4
5
6
7
8
4.35
4.40
4.45
4.50
4.55
4.60
4.65
4.70
4.75
4.80
1996 1998 2000 2002 2004 2006 2008 2010
MFOODPROCYFOODPROC
FOODPROC_LOG
foodproc
p e r c e n t
l o g of i n d ex
-2
-1
0
1
2
3
4
5
6
4.35
4.40
4.45
4.50
4.55
4.60
4.65
4.70
4.75
1996 1998 2000 2002 2004 2006 2008 2010
MSER V Y SERV SER V_LOG
serv
p e r c e n t
l o g of i n d ex
-4
-3
-2
-1
0
1
2
3
4.52
4.54
4.56
4.58
4.60
4.62
4.64
4.66
1996 1998 2000 2002 2004 2006 2008 2010
MIGOODSXE
YIGOODSXE
IGOODSXE_LOG
igoodsxe
p e r c e n t
l o g of i n d ex
Note: the graphs display the patterns of monthly inflation (“m” prefix), annual inflation (“y” prefix), on the left
y-axis, and the logarithm of the price index, on the right y-axis.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 26/37
24
Figure A.2. Intra-year pattern of monthly inflation; overall and by component
(1996:01-2010:12)
-1.2
-0.8
-0.4
0.0
0.4
0.8
1.2
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Means by Season
MCP00 by Season
-6
-4
-2
0
2
4
6
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Means by Season
MENRGY by Season
-2
-1
0
1
2
3
4
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Means by Season
MFOODUNP by Season
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Means by Season
MFOODPROC by Season
-0.8
-0.4
0.0
0.4
0.8
1.2
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Means by Season
MSERV by Season
-4
-3
-2
-1
0
1
2
3
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Means by Season
MIGOODSXE by Season
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 27/37
25
Tab le A.2.1: Desc rip tive sta tistic s – c risis eva luation period (2005:08-2010:12)
cp00 enrgy foodproc foodunp igoodsxe serv
Mean 1.94 4.71 2.56 2.10 0.66 2.09
Median 1.92 7.22 2.02 2.53 0.72 2.12
Maximum 4.06 17.01 7.20 4.63 1.18 2.81
Minimum -0.65 -14.25 0.35 -1.60 -0.12 1.18
Range 4.70 31.26 6.85 6.23 1.30 1.62
Std Dev 1.07 7.81 1.97 1.67 0.30 0.45
cp00 enrgy foodproc foodunp igoodsxe serv
Mean 0.17 0.34 0.21 0.18 0.10 0.16
Median 0.24 0.43 0.13 0.12 0.21 0.15
Maximum 1.11 3.53 1.18 1.47 2.48 0.89
Minimum -0.83 -4.90 -0.12 -1.23 -3.14 -0.75
Range 1.94 8.43 1.31 2.70 5.62 1.64Std Dev 0.38 1.75 0.25 0.58 1.23 0.43
Annual inflation (%)
Monthly inflation (%)
Source: own calculations based on Eurostat
Tab le A.2.2: Desc rip tive statistic s – p re-c risis eva lua tion period (2002:08-2007:12)
cp00 enrgy foodproc foodunp igoodsxe serv
Mean 2.17 5.39 2.75 1.86 0.74 2.46
Median 2.15 6.03 2.67 1.66 0.79 2.51
Maximum 3.07 14.93 5.10 4.63 1.33 3.35
Minimum 1.61 -2.04 1.45 -1.52 -0.02 1.84Range 1.45 16.97 3.64 6.15 1.35 1.51
Std Dev 0.29 4.52 0.84 1.46 0.31 0.33
cp00 enrgy foodproc foodunp igoodsxe serv
Mean 0.19 0.50 0.25 0.17 0.09 0.19
Median 0.25 0.45 0.15 0.23 0.16 0.15
Maximum 0.77 3.41 1.18 1.45 1.64 0.94
Minimum -0.60 -3.21 -0.11 -1.36 -2.06 -0.53
Range 1.37 6.62 1.30 2.81 3.70 1.47
Std Dev 0.29 1.45 0.28 0.61 0.94 0.38
Annual inflation (%)
Monthly inflation (%)
Source: own calculations based on Eurostat
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 28/37
26
Annex 2: Sta tionarity ana lysis
Tab le A.3 Aug mented Dickey-Fuller (ADF) test – Samp le: 1997:02-2012:02ADF-test cp00 enrgy foodproc foodunp igoodsxe serv
1% 5% 10%
log constant and 0 -3,09 -2,38 -2,16 -2,36 -7,99 -3,75 -4,01 -3,44 -3,14
trend 1 -3,4 -3,02 -2,34 -3,38 -14,17 -3,12
2 -3,17 -3,16 -2,65 -3,24 -10,85 -3
3 -2,85 -3,22 -2,79 -3,28 -9,82 -1,81
SIC -3,93 -3,02 -2,65 -3,86 -3,29 -3,81
AIC -3,33 -3,02 -3,03 -3,86 -2,54 -3,81
constant 0 -11,64 -9,71 -8,49 -9,78 -10,84 -15,46 -3,47 -2,88 -2,58
1 -9,55 -7,47 -5,48 -8,39 -14,15 -10,5
2 -8,85 -6,42 -4,66 -7,21 -13,2 -12,32
3 -8,58 -5,93 -4,23 -6,6 -26,81 -10,84
SIC -2,91 -9,71 -5,48 -2,55 -3,19 -0,85
AIC -3,21 -9,71 -2,89 -2,86 -3,19 -0,85
constant 0 -2,09 -2,08 -1,35 -2,11 -3,45 -1,83 -3,47 -2,88 -2,58
1 -2,65 -2,78 -2,19 -2,79 -3,72 -0,81
2 -3,06 -2,94 -2,91 -3,19 -2,97 -0,83
3 -2,92 -3,28 -2,92 -3,23 -2,7 -1,02
SIC -2,55 -2,77 -2,32 -2,79 -3,38 -2,67AIC -2,55 -2,77 -2,32 -2,67 -3,73 -2,21
d1d12 constant 0 -10,28 -9,84 -8,15 -9,94 -12,22 -20,67 -3,47 -2,88 -2,58
1 -7,1 -7,52 -5,24 -7,42 -11,23 -10,8
2 -6,64 -5,98 -4,89 -6,32 -9,36 -7,31
3 -5,24 -5,06 -3,78 -5,96 -9,2 -5,44
SIC -5,97 -6,16 -4,25 -9,94 -4,47 -20,67
AIC -5,97 -6,16 -4,25 -5,11 -4,47 -3,1
Critical values
Monthly
changes (d1)
Annual changes
(d12)
Deterministic
term
No. of lagged
differences
Philips-Peron (PP) test – Samp le: 1997:02-2012:02
PP-test cp00 enrgy foodproc foodunp igoodsxe serv
1% 5% 10%
log constant and -3,08 -2,86 -2,4 -2,89 -7,31 -3,68 -4,01 -3,44 -3,14
trend
Monthly constant -12,23 -9,7 -9,17 -9,81 -15,06 -20,43 -3,47 -2,88 -2,58
changes (d1)
Annual constant -2,75 -2,93 -2,44 -2,74 -3,68 -1,6 -3,47 -2,88 -2,58
changes (d12)
d1d12 constant -10,42 -9,99 -8,87 -9,94 -21,02 -19 -3,47 -2,88 -2,58
Deterministic
term
Critical values
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 29/37
27
Annex 3: Pre-selec tion o f exogenous va riab les
Tab le A.4: List of e xogeno us va riab les and abbrevia tions
oileur World, Energy, Oil, Crude Oil - North Sea (Brent), Dated, Close, EUR
oilusd World, Energy, Oil, Crude Oil - North Sea (Brent), Dated, Close, USDhwwieur Euro Zone, HWWI, Total Index, Average, EUR
hwwixeneur Euro Zone, HWWI, Total Index excluding Energy, Average, EUR
hwwiarm World, HWWI, Agricultural Raw Materials Index, Average, EUR
hwwicer World, HWWI, Cereals Index, Average, EUR
hwwicoal World, HWWI, Coal Index, Average, EUR
hwwicoil World, HWWI, Crude Oil Index, Average, EUR
hwwienrm World, HWWI, Energy Raw Materials Index, Average, EUR
hwwifood World, HWWI, Food Index, Average, EUR
hwwiind World, HWWI, Industrial Raw Materials Index, Average, EUR
hwwiiost World, HWWI, Iron Ore and Steel Scrap Index, Average, EUR
hwwinfm World, HWWI, Non-ferrous Metals Index, Average, EUR
hwwiseed World, HWWI, Oilseeds and Oils Index, Average, EUR
hwwitex World, HWWI, Textile Fibres Index, Average, EURhwwiexen World, HWWI, Total Index excluding Energy, Average, EUR
hwwi World, HWWI, Total Index, Average, EUR
eurusd Euro Zone, Spot Rates, EUR/USD, Close, USD
eer21 Euro Zone, ECB, EER-21 Nominal Effective Exchange Rate Index, Euro 16, EUR
eer41 Euro Zone, ECB, EER-41 Nominal Effective Exchange Rate Index, Euro 16, EUR
reer21 Euro Zone, ECB, EER-21 Real CPI Effective Exchange Rate Index, Euro 16, EUR
reer41 Euro Zone, ECB, EER-41 Real CPI Effective Exchange Rate Index, Euro 16, EUR
iptotencon Eurostat, Euro Zone, Production, Overall, NACE Rev.2 B_C_X_MIG_NRG, Total, excluding construction
and energy, Cal Adj, Index, EUR, 2005=100
iptotcon Eurostat, Euro Zone, Production, Overall, NACE Rev.2 B-D, Total, excluding construction, Cal Adj, Index,
EUR, 2005=100iptoten Eurostat, Euro Zone, Production, Overall, NACE Rev.2 B-D_F, Industry including construction, Cal Adj,
Index, EUR, 2005=100
ipcapg Eurostat, Euro Zone, Production, Aggregates, NACE Rev.2 MIG_CAG, Capital goods, Cal Adj, Index, EUR,
2005=100
ipconsg Eurostat, Euro Zone, Production, Aggregates, NACE Rev.2 MIG_COG, Consumer goods, Cal Adj, Index,
EUR, 2005=100
ipintgds Eurostat, Euro Zone, Production, Aggregates, NACE Rev.2 MIG_ING, Intermediate goods, Cal Adj, Index,
EUR, 2005=100
ipener Eurostat, Euro Zone, Production, Aggregates, NACE Rev.2 MIG_NRG_X_E, Energy, Cal Adj, Index, EUR,
2005=100
iocapg Eurostat, Euro Zone, New Orders, Aggregates, NACE Rev.2 MIG_CAG_ORD, Capital goods, Index, EUR,
2005=100ioconsd Eurostat, Euro Zone, New Orders, Aggregates, NACE Rev.2 MIG_DCOG_ORD, Consumer durables, Index,
EUR, 2005=100
iointg Eurostat, Euro Zone, New Orders, Aggregates, NACE Rev.2 MIG_ING_ORD, Intermediate goods, Index,
EUR, 2005=100
hhfinsit12 DG ECFIN, Euro Zone, Consumer Surveys, Financial situation of households over next 12 months, SA
hhecosit12 DG ECFIN, Euro Zone, Consumer Surveys, General economic situation over next 12 months, SA
majpur12 DG ECFIN, Euro Zone, Consumer Surveys, Major purchases over next 12 months, SA
pritre12 DG ECFIN, Euro Zone, Consumer Surveys, Price trends over next 12 months, SA
sav12 DG ECFIN, Euro Zone, Consumer Surveys, Savings over next 12 months, SA
uetot_lm Eurostat, Euro Zone, Unemployment, Overall, Total
uer_lm Eurostat, Euro Zone, Unemployment, Rate, Total
Commodity prices
Exchange rate
Industrial production
Industrial orders
Surveys
Labour market
Source: Ecowin
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 30/37
28
Tab le A.5: Lead ing ind ica to r ana lysis for head line infla tion (1997:02-2010:12)cp00 F-stat p-value R-bar2 F-stat p-value R-bar2 F-stat p-value R-bar2 F-stat p-value R-bar2
dln_oileur 8.66 0.00 0.63 5.61 0.00 0.56 3.49 0.00 0.50 1.43 0.15 0.41
dln_oilusd 8.02 0.00 0.64 5.67 0.00 0.58 3.40 0.00 0.51 1.56 0.11 0.43
dln_hwwieur 9.60 0.00 0.67 5.71 0.00 0.58 2.95 0.00 0.49 1.14 0.34 0.41
dln_hwwixeneur 3.79 0.00 0.52 3.42 0.00 0.51 1.50 0.13 0.43 0.66 0.80 0.38dln_hwwiarm 1.86 0.04 0.45 2.17 0.01 0.46 1.37 0.18 0.42 1.02 0.44 0.40
dln_hwwicer 3.39 0.00 0.51 3.31 0.00 0.51 1.55 0.11 0.43 1.02 0.44 0.40
dln_hwwicoal 4.02 0.00 0.53 3.90 0.00 0.53 1.48 0.13 0.43 1.88 0.04 0.45
dln_hwwicoil 7.78 0.00 0.63 5.25 0.00 0.57 3.16 0.00 0.50 1.06 0.40 0.40
dln_hwwienrm 8.22 0.00 0.64 5.55 0.00 0.58 3.21 0.00 0.50 1.14 0.34 0.41
dln_hwwifood 2.80 0.00 0.49 2.82 0.00 0.49 1.41 0.16 0.42 0.93 0.52 0.40
dln_hwwiind 3.13 0.00 0.50 2.54 0.00 0.47 1.38 0.18 0.42 0.88 0.58 0.39
dln_hwwiiost 3.38 0.00 0.51 3.28 0.00 0.50 3.32 0.00 0.51 1.38 0.18 0.42
dln_hwwinfm 3.01 0.00 0.49 1.95 0.03 0.45 1.10 0.36 0.41 0.66 0.79 0.38
dln_hwwiseed 1.80 0.05 0.44 1.95 0.03 0.45 1.22 0.27 0.41 1.27 0.24 0.42
dln_hwwitex 1.81 0.05 0.44 1.84 0.04 0.44 1.24 0.26 0.41 0.54 0.89 0.37
dln_hwwiexen 3.31 0.00 0.51 2.95 0.00 0.49 1.24 0.26 0.41 0.66 0.80 0.38
dln_hwwi 9.02 0.00 0.65 5.70 0.00 0.58 3.01 0.00 0.50 1.08 0.38 0.41
dln_eurusd 1.35 0.19 0.42 1.64 0.08 0.43 0.79 0.67 0.39 0.91 0.54 0.40
dln_eer21 1.13 0.34 0.41 1.43 0.16 0.42 0.61 0.84 0.38 1.01 0.45 0.40
dln_eer41 1.11 0.36 0.41 1.40 0.17 0.42 0.63 0.82 0.38 0.91 0.54 0.40dln_reer21 1.20 0.28 0.41 1.42 0.16 0.42 0.58 0.87 0.38 0.93 0.52 0.40
dln_reer41 1.15 0.33 0.41 1.37 0.18 0.42 0.58 0.86 0.38 0.88 0.58 0.39
dln_iptotencon 5.57 0.00 0.58 5.62 0.00 0.58 4.43 0.00 0.55 4.41 0.00 0.55
dln_iptotcon 6.00 0.00 0.59 6.29 0.00 0.60 5.15 0.00 0.57 5.11 0.00 0.58
dln_iptoten 6.23 0.00 0.60 6.35 0.00 0.60 5.24 0.00 0.57 5.41 0.00 0.58
dln_ipcapg 6.44 0.00 0.60 6.34 0.00 0.60 5.41 0.00 0.58 4.91 0.00 0.57
dln_ipconsg 4.38 0.00 0.54 4.42 0.00 0.54 4.94 0.00 0.56 5.02 0.00 0.57
dln_ipintgds 5.87 0.00 0.59 5.56 0.00 0.58 4.41 0.00 0.55 4.61 0.00 0.56
dln_ipener 4.32 0.00 0.54 4.77 0.00 0.56 4.60 0.00 0.55 5.23 0.00 0.58
dln_iocapg 5.26 0.00 0.57 5.21 0.00 0.57 2.80 0.00 0.49 3.10 0.00 0.50
dln_ioconsd 4.73 0.00 0.55 4.42 0.00 0.54 3.04 0.00 0.50 2.97 0.00 0.50
dln_iointg 4.72 0.00 0.55 4.99 0.00 0.56 2.39 0.01 0.47 2.28 0.01 0.47
plv_hhfinsit12 0.62 0.83 0.38 0.63 0.82 0.38 1.01 0.45 0.40 0.82 0.64 0.39plv_hhecosit12 0.82 0.63 0.39 0.83 0.63 0.39 1.38 0.18 0.42 1.08 0.39 0.41
plv_majpur12 1.62 0.09 0.43 1.66 0.08 0.43 1.07 0.39 0.40 1.11 0.35 0.41
plv_pritre12 1.79 0.05 0.44 1.73 0.06 0.44 1.68 0.07 0.44 0.75 0.72 0.39
dlv_sav12 1.32 0.21 0.42 1.31 0.22 0.42 1.36 0.19 0.42 1.56 0.11 0.43
dln_uetot_lm 3.40 0.00 0.51 2.54 0.00 0.47 2.98 0.00 0.50 4.28 0.00 0.55
dln_uer_lm 4.35 0.00 0.54 2.75 0.00 0.48 3.02 0.00 0.50 4.08 0.00 0.54
Industrial production
Industrial orders
Surveys
Labour market
k=0 k=1 k=6 k=12
Commodity prices
Exchange rate
Note: the adjusted coefficient of determination of the (restricted) AR(12) model is 0.39. "dln" is the first
difference of the logarithmic series; "dlv" is the first difference of the original series; "plv" is the percentage
change of the original series.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 31/37
29
Annex 4: Rec ursive Root Mea n Squared Forec ast Error
The forecast accuracy of a model m can be measured by the standard recursive Root Mean
Squared Forecast Error (RMSFE) in a simulated out-of-sample forecast exercise. The RMSFE
is obtained by first splitting the sample (of length N ) into two parts: the model is fitted to the
first subsample (of length N-n), running up to month t , and the series is then forecasted "out-of-sample". The forecast errors – the difference between the expectation of annual inflation at
time t+h conditional on the information at time t ,21
and the actual value of annual inflation –
are then computed over the second (shorter) sub-sample (of length n) and stored.22
The
maximum number of comparable forecasts thus corresponds to the number of remaining
observations in the second sub-sample. For a given forecast horizon, the RMSFE is given by
the following expression:
( )2 / 1
2
/ ˆ)( ÷øö
çè
æ -= å ++ T h RMSFE T
t ht t
mmht p p ,
Where mht +
p ̂ is the annual inflation forecast generated by candidate model m, at an h-month
horizon, conditional on the information or assumptions at time t . π t+h is the actual inflation
rate at time t+h, while T represents the number of observations over which the model is
evaluated.
The first sub-sample is recursively extended by one observation, and the estimation and
comparison procedure above is repeated. While the first sub-sample thus gradually
approaches the size of the full sample, the maximum comparable forecast horizon is reduced
by one period at each step and thus approaches zero. More generally, for an h-step-ahead
forecast, there will be n-h+1 possible forecast observations.23
If a model is to be evaluated upto h steps ahead, the last recursive estimation should yield h comparable forecasts, implying
that the last h available observations will not be used for the recursive estimation but only for
evaluation purposes. This also implies that the same number of observations, T (=n- hmax+1),
is used to compute the RMSFE at each forecast horizon. In this paper, the evaluation period
(T ) is set to 48 months (i.e. 4 years), and the maximum forecast horizon (hmax) is set to 18
months.
21 If the model contains exogenous variables, the forecast needs to be conditioned on a specific path assumed for
the latter beyond time t.22
Models are compared on how accurately they predict annual inflation, which is the natural yardstick for
inflation. This implies that forecasts obtained from models that are based on other transformations of the original
price level series first need to be transformed into annual price changes.23 Starting with a given sub-sample, one will obtain n possible one-step-ahead forecasts, n-1 two-step ahead
forecasts etc. and one n-step-ahead forecast.
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 32/37
Annex 5: Results
Tab le.A.6: RMSFEs fo r eva lua tion sam ple inc lud ing c risis
Horizon: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks
ECB 1,03 1,08 1,12 1,15 1,17 1,18 1,18 1,19 1,19 1,19 1,19 1,19 1,19 1,19 1,19 1,18 1,18 1,19 1,12 1,19 1,19 1,17
Naive 0,31 0,51 0,70 0,87 1,04 1,18 1,31 1,44 1,57 1,68 1,77 1,84 1,89 1,90 1,90 1,89 1,86 1,84 0,77 1,60 1,88 1,42
AR(1) 0,37 0,57 0,72 0,81 0,87 0,90 1,02 1,12 1,19 1,22 1,23 1,24 1,24 1,24 1,24 1,24 1,23 1,23 0,71 1,17 1,24 1,04
ARMA(1,1) 0,37 0,56 0,70 0,79 0,85 0,89 1,01 1,10 1,17 1,19 1,20 1,22 1,22 1,22 1,22 1,21 1,21 1,20 0,69 1,15 1,21 1,02
AR(12) 0,31 0,53 0,74 0,91 1,06 1,17 1,29 1,40 1,49 1,57 1,64 1,69 1,70 1,71 1,69 1,68 1,64 1,61 0,79 1,51 1,67 1,32
Seas. Dummies 0,26 0,41 0,54 0,64 0,73 0,81 0,91 1,00 1,08 1,14 1,19 1,24 1,24 1,24 1,24 1,23 1,23 1,23 0,57 1,09 1,23 0,96
Univariate (AIC) AR MA
U.1 10 6 0,31 0,48 0,66 0,81 0,94 1,01 1,11 1,20 1,26 1,33 1,36 1,39 1,38 1,38 1,35 1,33 1,31 1,29 0,70 1,27 1,34 1,11
U.2 8 9 0,27 0,44 0,60 0,71 0,83 0,92 1,05 1,19 1,30 1,40 1,48 1,56 1,57 1,56 1,55 1,52 1,47 1,40 0,63 1,33 1,51 1,16
U.3 10 9 0,30 0,50 0,70 0,85 1,00 1,12 1,23 1,34 1,43 1,50 1,54 1,57 1,56 1,55 1,53 1,52 1,49 1,46 0,75 1,43 1,52 1,23
U.4 5 6 0,26 0,42 0,57 0,66 0,74 0,82 0,91 1,00 1,09 1,16 1,21 1,26 1,27 1,27 1,27 1,27 1,25 1,24 0,58 1,11 1,26 0,98
Univariate (SIC) AR MA
U.1 10 6 0,31 0,48 0,66 0,81 0,94 1,01 1,11 1,20 1,26 1,33 1,36 1,39 1,38 1,38 1,35 1,33 1,31 1,29 0,70 1,27 1,34 1,11U.2 7 5 0,24 0,42 0,56 0,67 0,78 0,86 0,97 1,08 1,16 1,24 1,31 1,37 1,38 1,38 1,38 1,37 1,35 1,32 0,59 1,19 1,36 1,05
U.3 0 0 0,29 0,48 0,66 0,80 0,95 1,07 1,19 1,31 1,43 1,53 1,62 1,69 1,72 1,73 1,73 1,72 1,69 1,66 0,71 1,46 1,71 1,29
U.4 5 6 0,26 0,42 0,57 0,66 0,74 0,82 0,91 1,00 1,09 1,16 1,21 1,26 1,27 1,27 1,27 1,27 1,25 1,24 0,58 1,11 1,26 0,98
ADL Specification Y-lags X-lags
oileur B.3 7 6 0,19 0,28 0,36 0,41 0,48 0,52 0,61 0,70 0,79 0,87 0,93 1,00 1,02 1,04 1,05 1,06 1,04 1,05 0,37 0,82 1,04 0,74
oilusd B.3 7 9 0,19 0,27 0,34 0,38 0,44 0,48 0,56 0,64 0,72 0,80 0,86 0,92 0,93 0,95 0,95 0,96 0,94 0,94 0,35 0,75 0,95 0,68
hwwieur B.3 7 6 0,19 0,27 0,35 0,38 0,43 0,45 0,53 0,62 0,70 0,77 0,82 0,88 0,90 0,92 0,93 0,93 0,93 0,94 0,34 0,72 0,92 0,66
hwwixeneur B .1 6 2 0,31 0,44 0,56 0,63 0,68 0,71 0,79 0,87 0,94 0,99 1,03 1,08 1,09 1,12 1,13 1,13 1,14 1,14 0,55 0,95 1,12 0,88
hwwiarm B.3 6 0 0,25 0,42 0,56 0,67 0,77 0,84 0,94 1,04 1,14 1,21 1,27 1,32 1,32 1,32 1,31 1,29 1,27 1,25 0,59 1,16 1,29 1,01
hwwicer B.1 8 9 0,32 0,44 0,57 0,64 0,69 0,70 0,75 0,81 0,84 0,86 0,86 0,87 0,87 0,87 0,89 0,90 0,90 0,89 0,56 0,83 0,89 0,76
hwwicoal B.1 6 19 0,33 0,44 0,53 0,60 0,65 0,66 0,75 0,84 0,90 0,94 0,96 0,98 0,97 0,96 0,95 0,95 0,95 0,94 0,54 0,90 0,95 0,79
hwwicoil B.3 7 6 0,19 0,28 0,36 0,41 0,47 0,52 0,60 0,69 0,78 0,86 0,92 0,99 1,01 1,03 1,04 1,05 1,03 1,04 0,37 0,80 1,03 0,74
hwwienrm B.3 7 9 0,19 0,28 0,36 0,40 0,47 0,51 0,58 0,67 0,76 0,85 0,91 0,98 1,00 1,02 1,02 1,03 1,02 1,02 0,37 0,79 1,02 0,73
hwwifood B.1 8 9 0,31 0,46 0,61 0,69 0,75 0,77 0,84 0,92 0,96 0,98 0,99 1,00 1,00 1,00 1,00 1,01 1,01 1,00 0,60 0,95 1,00 0,85
hwwiind B.3 6 0 0,24 0,39 0,52 0,61 0,71 0,78 0,88 0,98 1,08 1,16 1,22 1,28 1,29 1,29 1,29 1,28 1,27 1,25 0,54 1,10 1,28 0,97
hwwiiost B.1 7 16 0,32 0,48 0,57 0,63 0,66 0,68 0,76 0,83 0,87 0,89 0,92 0,96 0,99 1,02 1,03 1,05 1,06 1,06 0,55 0,87 1,03 0,82
hwwinfm B.3 6 0 0,24 0,38 0,51 0,60 0,70 0,77 0,88 0,98 1,08 1,16 1,23 1,29 1,30 1,30 1,29 1,29 1,27 1,26 0,53 1,10 1,29 0,97
hwwiseed B.4 6 9 0,26 0,42 0,55 0,63 0,71 0,76 0,83 0,92 1,01 1,07 1,11 1,15 1,14 1,14 1,13 1,12 1,10 1,09 0,55 1,01 1,12 0,90
hwwiexen B.3 6 2 0,24 0,38 0,51 0,59 0,68 0,73 0,83 0,92 1,01 1,09 1,14 1,20 1,21 1,22 1,22 1,21 1,21 1,20 0,52 1,03 1,21 0,92
hwwi B.3 7 6 0,19 0,28 0,35 0,38 0,44 0,47 0,55 0,63 0,72 0,79 0,85 0,91 0,93 0,96 0,96 0,97 0,97 0,97 0,35 0,74 0,96 0,68
eurusd B.3 6 n.a. 0,25 0,41 0,56 0,66 0,76 0,83 0,93 1,03 1,12 1,19 1,25 1,30 1,29 1,29 1,28 1,27 1,25 1,24 0,58 1,14 1,27 0,99
iptotencon B.1 11 9 0,27 0,44 0,60 0,71 0,81 0,85 0,90 0,92 0,94 0,95 0,97 0,99 0,97 0,98 0,97 0,95 0,91 0,88 0,61 0,94 0,94 0,83
iptotcon B.2 6 8 0,25 0,42 0,58 0,70 0,81 0,87 0,94 0,98 1,03 1,06 1,08 1,09 1,06 1,05 1,01 0,97 0,92 0,89 0,61 1,03 0,98 0,87
iptoten B.3 6 1 0,24 0,40 0,54 0,65 0,75 0,82 0,91 1,00 1,09 1,16 1,21 1,25 1,24 1,24 1,22 1,20 1,16 1,14 0,57 1,10 1,20 0,96
ipcapg B.2 8 8 0,25 0,42 0,57 0,68 0,78 0,83 0,89 0,92 0,95 0,98 0,99 1,00 0,99 0,99 0,97 0,95 0,92 0,90 0,59 0,96 0,95 0,83
ipconsg B.3 6 6 0,25 0,41 0,56 0,65 0,77 0,84 0,95 1,05 1,14 1,22 1,29 1,34 1,34 1,34 1,33 1,31 1,27 1,25 0,58 1,17 1,31 1,02
ipintgds B.2 11 10 0,26 0,43 0,58 0,68 0,76 0,80 0,87 0,95 1,05 1,14 1,23 1,31 1,33 1,35 1,34 1,32 1,26 1,22 0,59 1,09 1,30 0,99
ipener B.3 6 13 0,26 0,44 0,59 0,68 0,78 0,82 0,88 0,95 1,04 1,11 1,16 1,19 1,18 1,17 1,15 1,13 1,11 1,09 0,59 1,05 1,14 0,93
iocapg B.2 0 7 0,25 0,38 0,50 0,58 0,67 0,71 0,76 0,81 0,87 0,91 0,94 0,98 0,98 0,97 0,94 0,92 0,88 0,86 0,52 0,88 0,93 0,77
ioconsd B.1 8 12 0,28 0,48 0,66 0,83 0,96 1,04 1,14 1,20 1,24 1,27 1,31 1,34 1,34 1,34 1,33 1,33 1,32 1,31 0,71 1,25 1,33 1,10
iointg B.1 6 19 0,28 0,42 0,56 0,64 0,68 0,70 0,74 0,76 0,80 0,84 0,87 0,89 0,89 0,90 0,92 0,94 0,95 0,97 0,55 0,81 0,93 0,76
hhfinsit12 B.3 6 3 0,25 0,41 0,56 0,66 0,76 0,83 0,93 1,03 1,13 1,21 1,27 1,32 1,32 1,32 1,31 1,30 1,27 1,26 0,58 1,15 1,30 1,01
hhecosit12 B.3 6 n.a. 0,25 0,41 0,56 0,66 0,76 0,83 0,93 1,03 1,12 1,19 1,25 1,30 1,29 1,29 1,28 1,27 1,25 1,24 0,58 1,14 1,27 0,99
majpur12 B.1 8 3 0,34 0,52 0,69 0,82 0,92 0,97 1,07 1,15 1,20 1,23 1,24 1,25 1,23 1,22 1,22 1,22 1,22 1,20 0,71 1,19 1,22 1,04pritre12 B.3 6 0 0,25 0,41 0,55 0,64 0,74 0,80 0,90 0,99 1,09 1,16 1,23 1,29 1,32 1,34 1,36 1,37 1,37 1,37 0,56 1,11 1,35 1,01
sav12 B.3 6 1 0,24 0,40 0,54 0,63 0,71 0,77 0,85 0,94 1,02 1,09 1,13 1,17 1,16 1,16 1,14 1,13 1,11 1,10 0,55 1,03 1,13 0,91
uetot_lm B.3 6 0 0,24 0,38 0,50 0,58 0,67 0,72 0,80 0,87 0,95 1,01 1,06 1,09 1,08 1,08 1,08 1,08 1,06 1,05 0,51 0,96 1,07 0,85
uer_lm B.3 6 0 0,23 0,37 0,50 0,58 0,66 0,72 0,80 0,87 0,95 1,01 1,06 1,09 1,08 1,09 1,09 1,09 1,07 1,06 0,51 0,96 1,08 0,85
Agg-AR(1) 0,36 0,57 0,73 0,83 0,88 0,89 1,00 1,11 1,19 1,24 1,24 1,24 1,24 1,24 1,24 1,24 1,23 1,23 0,71 1,17 1,24 1,04
Agg-ARMA(1,1) 0,36 0,58 0,73 0,81 0,86 0,85 0,96 1,08 1,18 1,23 1,25 1,24 1,24 1,25 1,25 1,25 1,24 1,23 0,70 1,16 1,24 1,03
Agg-AR(12) 0,23 0,38 0,54 0,68 0,82 0,93 1,04 1,14 1,23 1,31 1,39 1,47 1,49 1,50 1,50 1,49 1,47 1,45 0,60 1,26 1,48 1,12
Agg-Seas. Dummies 0,27 0,43 0,56 0,64 0,73 0,80 0,90 1,00 1,09 1,15 1,19 1,24 1,24 1,24 1,24 1,23 1,23 1,23 0,57 1,09 1,23 0,97
Agg-ARMA (identical) 0,21 0,36 0,52 0,62 0,72 0,82 0,92 1,00 1,07 1,15 1,21 1,28 1,31 1,32 1,32 1,32 1,31 1,30 0,54 1,11 1,31 0,99
Agg-ARMA (component-specific) 0,20 0,32 0,45 0,54 0,65 0,74 0,82 0,92 0,99 1,07 1,13 1,19 1,20 1,21 1,21 1,20 1,20 1,20 0,48 1,02 1,20 0,90
Agg-oilusd (identical) 0,16 0,23 0,31 0,36 0,44 0,49 0,56 0,62 0,67 0,73 0,79 0,84 0,86 0,88 0,89 0,90 0,91 0,91 0,33 0,70 0,89 0,64
Agg-oilusd (component-specific) 0,16 0,21 0,29 0,34 0,42 0,47 0,53 0,60 0,65 0,70 0,75 0,80 0,82 0,85 0,86 0,87 0,89 0,89 0,32 0,67 0,86 0,62
Agg-multivariate (component-specific) 0,16 0,21 0,28 0,31 0,36 0,38 0,39 0,42 0,45 0,48 0,50 0,53 0,53 0,53 0,53 0,53 0,53 0,54 0,28 0,46 0,53 0,42
Direct forecasts
Indirect forecasts
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 33/37
31
Tab le.A.7: RMSFEs for p re-c risis eva lua tion sam ple
Horizon: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks
ECB 0,34 0,34 0,34 0,34 0,33 0,33 0,33 0,32 0,31 0,31 0,31 0,31 0,31 0,31 0,31 0,33 0,36 0,40 0,34 0,32 0,34 0,33
Naive 0,18 0,24 0,30 0,33 0,34 0,34 0,34 0,35 0,34 0,35 0,37 0,40 0,39 0,38 0,36 0,35 0,35 0,36 0,29 0,36 0,36 0,34
AR(1) 0,29 0,41 0,49 0,51 0,49 0,46 0,52 0,58 0,58 0,55 0,48 0,38 0,37 0,37 0,37 0,38 0,41 0,44 0,44 0,52 0,39 0,45
ARMA(1,1) 0,29 0,40 0,47 0,49 0,47 0,44 0,49 0,55 0,55 0,52 0,45 0,36 0,36 0,36 0,35 0,36 0,40 0,43 0,42 0,49 0,38 0,43
AR(12) 0,20 0,27 0,32 0,35 0,35 0,35 0,35 0,36 0,36 0,35 0,34 0,33 0,33 0,34 0,34 0,35 0,37 0,39 0,31 0,35 0,35 0,34
Seas. Dummies 0,18 0,27 0,33 0,34 0,34 0,33 0,36 0,40 0,41 0,41 0,40 0,41 0,41 0,40 0,40 0,41 0,44 0,47 0,30 0,40 0,42 0,37
Univariate (AIC) AR MA
U.1 10 10 0,23 0,29 0,35 0,37 0,38 0,38 0,41 0,41 0,41 0,42 0,39 0,37 0,35 0,36 0,37 0,37 0,40 0,43 0,34 0,40 0,38 0,37
U.2 7 8 0,18 0,25 0,32 0,35 0,35 0,36 0,37 0,38 0,41 0,43 0,43 0,43 0,42 0,41 0,42 0,42 0,44 0,45 0,30 0,41 0,43 0,38
U.3 10 11 0,20 0,26 0,33 0,37 0,39 0,38 0,41 0,44 0,42 0,42 0,42 0,42 0,43 0,46 0,45 0,44 0,44 0,47 0,32 0,42 0,45 0,40
U.4 2 11 0,18 0,27 0,35 0,39 0,41 0,42 0,45 0,50 0,53 0,53 0,53 0,51 0,49 0,48 0,48 0,45 0,43 0,42 0,34 0,51 0,46 0,44
Univariate (SIC) AR MA
U.1 9 7 0,25 0,32 0,39 0,41 0,43 0,42 0,42 0,44 0,42 0,43 0,41 0,38 0,37 0,38 0,37 0,36 0,39 0,41 0,37 0,42 0,38 0,39
U.2 2 2 0,18 0,26 0,32 0,34 0,33 0,33 0,36 0,39 0,39 0,39 0,39 0,39 0,40 0,40 0,40 0,39 0,41 0,44 0,29 0,39 0,40 0,36
U.3 0 0 0,19 0,26 0,32 0,35 0,35 0,34 0,36 0,39 0,38 0,38 0,37 0,36 0,36 0,36 0,36 0,36 0,36 0,37 0,30 0,37 0,36 0,35
U.4 2 2 0,18 0,25 0,31 0,32 0,31 0,30 0,31 0,35 0,36 0,37 0,34 0,34 0,32 0,34 0,34 0,37 0,40 0,42 0,28 0,35 0,37 0,33
ADL Specification Y-lags X-lags
oileur B.4 7 1 0,16 0,19 0,24 0,26 0,26 0,25 0,26 0,28 0,28 0,29 0,29 0,29 0,28 0,27 0,27 0,27 0,29 0,31 0,23 0,28 0,28 0,26 oilusd B.4 7 1 0,15 0,19 0,23 0,24 0,24 0,24 0,24 0,26 0,26 0,27 0,27 0,25 0,24 0,23 0,22 0,22 0,24 0,26 0,22 0,26 0,23 0,24
hwwieur B.3 6 1 0,17 0,21 0,26 0,28 0,27 0,27 0,28 0,31 0,31 0,31 0,31 0,30 0,29 0,29 0,27 0,27 0,28 0,30 0,24 0,30 0,28 0,28
hwwixeneur B.2 4 1 0,19 0,24 0,29 0,32 0,34 0,33 0,34 0,35 0,33 0,32 0,32 0,32 0,32 0,32 0,32 0,32 0,34 0,37 0,29 0,33 0,33 0,32
hwwiarm B.2 6 1 0,19 0,23 0,28 0,30 0,32 0,31 0,32 0,34 0,33 0,32 0,33 0,35 0,35 0,35 0,35 0,35 0,37 0,39 0,27 0,33 0,36 0,32
hwwicer B.4 6 6 0,18 0,23 0,28 0,30 0,32 0,33 0,37 0,41 0,44 0,47 0,50 0,53 0,54 0,54 0,54 0,53 0,54 0,55 0,28 0,45 0,54 0,42
hwwicoal B.2 4 4 0,21 0,25 0,29 0,31 0,32 0,31 0,34 0,37 0,37 0,38 0,39 0,39 0,39 0,40 0,39 0,39 0,40 0,41 0,28 0,37 0,40 0,35
hwwicoil B .4 7 1 0,16 0,20 0,24 0,26 0,26 0,26 0,26 0,28 0,29 0,30 0,30 0,30 0,29 0,29 0,29 0,29 0,31 0,33 0,23 0,29 0,30 0,27
hwwienrm B.4 7 1 0,16 0,20 0,24 0,26 0,25 0,25 0,25 0,28 0,28 0,29 0,29 0,29 0,28 0,29 0,28 0,28 0,30 0,32 0,23 0,28 0,29 0,27
hwwifood B.4 11 6 0,18 0,20 0,24 0,24 0,24 0,24 0,27 0,31 0,33 0,34 0,35 0,35 0,35 0,36 0,36 0,36 0,36 0,38 0,22 0,32 0,36 0,30
hwwiind B.4 2 8 0,17 0,23 0,28 0,30 0,31 0,31 0,32 0,34 0,33 0,33 0,33 0,35 0,35 0,34 0,34 0,34 0,36 0,39 0,27 0,33 0,35 0,32
hwwiiost B .2 4 1 0,19 0,26 0,31 0,33 0,34 0,34 0,34 0,33 0,30 0,29 0,28 0,29 0,29 0,29 0,28 0,29 0,32 0,35 0,29 0,31 0,30 0,30
hwwinfm B.4 6 0 0,17 0,23 0,27 0,30 0,30 0,30 0,31 0,32 0,31 0,31 0,31 0,31 0,31 0,29 0,28 0,30 0,35 0,40 0,26 0,31 0,32 0,30
hwwiseed B.4 11 6 0,18 0,21 0,25 0,25 0,25 0,24 0,27 0,31 0,34 0,35 0,35 0,37 0,36 0,37 0,37 0,38 0,39 0,40 0,23 0,33 0,38 0,31
hwwitex B.2 4 n.a. 0,20 0,25 0,30 0,32 0,33 0,33 0,34 0,35 0,34 0,34 0,34 0,34 0,35 0,35 0,35 0,36 0,38 0,41 0,29 0,34 0,36 0,33
hwwiexen B.2 4 1 0,19 0,24 0,29 0,32 0,33 0,33 0,33 0,34 0,33 0,31 0,31 0,31 0,32 0,32 0,32 0,33 0,35 0,38 0,28 0,32 0,34 0,31
hwwi B.3 6 1 0,17 0,21 0,26 0,27 0,27 0,27 0,27 0,30 0,31 0,31 0,30 0,29 0,28 0,28 0,28 0,27 0,28 0,31 0,24 0,30 0,28 0,27
eurusd B.4 6 n.a. 0,17 0,22 0,27 0,30 0,31 0,31 0,33 0,35 0,36 0,36 0,38 0,39 0,39 0,39 0,39 0,40 0,43 0,46 0,26 0,36 0,41 0,35
eer21 B.4 6 n.a. 0,17 0,22 0,27 0,30 0,31 0,31 0,33 0,35 0,36 0,36 0,38 0,39 0,39 0,39 0,39 0,40 0,43 0,46 0,26 0,36 0,41 0,35
eer41 B.2 4 1 0,19 0,24 0,29 0,32 0,33 0,33 0,34 0,35 0,33 0,32 0,33 0,33 0,33 0,33 0,33 0,34 0,36 0,40 0,28 0,33 0,35 0,32
reer21 B.4 6 n.a. 0,17 0,22 0,27 0,30 0,31 0,31 0,33 0,35 0,36 0,36 0,38 0,39 0,39 0,39 0,39 0,40 0,43 0,46 0,26 0,36 0,41 0,35
reer41 B.2 4 1 0,20 0,25 0,30 0,32 0,34 0,33 0,35 0,36 0,34 0,34 0,34 0,35 0,35 0,35 0,35 0,36 0,39 0,42 0,29 0,35 0,37 0,33
iptotencon B.2 11 11 0,17 0,24 0,28 0,30 0,30 0,29 0,30 0,32 0,34 0,35 0,35 0,36 0,36 0,36 0,35 0,35 0,38 0,41 0,27 0,34 0,37 0,32
iptotcon B.2 6 13 0,16 0,22 0,26 0,28 0,29 0,29 0,30 0,31 0,32 0,32 0,34 0,36 0,35 0,35 0,36 0,37 0,41 0,45 0,25 0,33 0,38 0,32
iptoten B.2 6 16 0,18 0,25 0,32 0,36 0,37 0,37 0,36 0,36 0,33 0,33 0,35 0,39 0,39 0,39 0,38 0,38 0,41 0,45 0,31 0,35 0,40 0,35
ipcapg B.2 6 11 0,17 0,22 0,27 0,29 0,30 0,30 0,31 0,32 0,32 0,33 0,35 0,36 0,36 0,35 0,36 0,38 0,42 0,46 0,26 0,33 0,39 0,33
ipconsg B.2 4 4 0,17 0,23 0,28 0,31 0,32 0,32 0,34 0,35 0,35 0,35 0,36 0,37 0,37 0,37 0,37 0,39 0,42 0,45 0,27 0,35 0,39 0,34
ipintgds B.2 11 17 0,18 0,25 0,29 0,31 0,31 0,30 0,31 0,34 0,36 0,38 0,39 0,40 0,39 0,39 0,38 0,38 0,39 0,43 0,27 0,36 0,39 0,34
ipener B.2 10 15 0,19 0,25 0,29 0,31 0,31 0,30 0,31 0,34 0,35 0,35 0,35 0,33 0,30 0,30 0,29 0,30 0,33 0,36 0,27 0,34 0,31 0,31
iocapg B.2 10 11 0,19 0,25 0,30 0,31 0,31 0,29 0,30 0,33 0,35 0,37 0,37 0,38 0,37 0,37 0,37 0,36 0,36 0,38 0,28 0,35 0,37 0,33
ioconsd B.2 4 11 0,18 0,24 0,29 0,31 0,32 0,32 0,34 0,35 0,35 0,36 0,36 0,37 0,36 0,36 0,36 0,38 0,40 0,43 0,28 0,35 0,38 0,34
iointg B.2 4 12 0,18 0,23 0,27 0,29 0,30 0,30 0,32 0,34 0,33 0,35 0,36 0,37 0,38 0,37 0,37 0,37 0,40 0,42 0,26 0,35 0,38 0,33
hhfinsit12 B.2 4 n.a. 0,20 0,25 0,30 0,32 0,33 0,33 0,34 0,35 0,34 0,34 0,34 0,34 0,35 0,35 0,35 0,36 0,38 0,41 0,29 0,34 0,36 0,33
hhecosit12 B.2 4 n.a. 0,20 0,25 0,30 0,32 0,33 0,33 0,34 0,35 0,34 0,34 0,34 0,34 0,35 0,35 0,35 0,36 0,38 0,41 0,29 0,34 0,36 0,33
majpur12 B.2 4 11 0,21 0,27 0,33 0,35 0,37 0,37 0,38 0,39 0,38 0,38 0,39 0,41 0,42 0,42 0,44 0,45 0,46 0,47 0,32 0,39 0,44 0,38
pritre12 B.2 6 0 0,20 0,24 0,27 0,28 0,29 0,28 0,30 0,31 0,30 0,29 0,29 0,29 0,30 0,30 0,31 0,31 0,33 0,34 0,26 0,30 0,31 0,29sav12 B.2 4 0 0,19 0,25 0,30 0,32 0,33 0,33 0,34 0,35 0,34 0,33 0,34 0,34 0,34 0,35 0,35 0,36 0,38 0,40 0,28 0,34 0,36 0,33
uetot_lm B.4 6 n.a. 0,17 0,22 0,27 0,30 0,31 0,31 0,33 0,35 0,36 0,36 0,38 0,39 0,39 0,39 0,39 0,40 0,43 0,46 0,26 0,36 0,41 0,35
uer_lm B.4 6 n.a. 0,17 0,22 0,27 0,30 0,31 0,31 0,33 0,35 0,36 0,36 0,38 0,39 0,39 0,39 0,39 0,40 0,43 0,46 0,26 0,36 0,41 0,35
Aggregation
Agg-AR(1) 0,29 0,43 0,53 0,57 0,54 0,47 0,53 0,61 0,63 0,61 0,52 0,39 0,37 0,36 0,36 0,37 0,40 0,43 0,47 0,55 0,38 0,47
Agg-ARMA(1,1) 0,29 0,44 0,54 0,56 0,54 0,45 0,52 0,61 0,64 0,61 0,53 0,37 0,32 0,31 0,30 0,31 0,33 0,36 0,47 0,55 0,32 0,45
Agg-AR(12) 0,16 0,24 0,30 0,34 0,34 0,34 0,34 0,37 0,37 0,36 0,34 0,35 0,34 0,33 0,32 0,31 0,33 0,36 0,29 0,35 0,33 0,32
Agg-Seas. Dummies 0,20 0,29 0,36 0,36 0,34 0,32 0,36 0,41 0,43 0,42 0,40 0,39 0,39 0,39 0,39 0,39 0,42 0,45 0,31 0,40 0,40 0,37
Agg-ARMA (identical) 0,16 0,23 0,31 0,35 0,37 0,38 0,39 0,44 0,46 0,49 0,52 0,55 0,54 0,53 0,52 0,53 0,58 0,62 0,30 0,48 0,55 0,44
Agg-ARMA (component-specific) 0,15 0,23 0,30 0,33 0,33 0,34 0,35 0,36 0,35 0,35 0,35 0,37 0,35 0,33 0,31 0,31 0,35 0,38 0,28 0,36 0,34 0,33
Agg-oilusd (identical) 0,14 0,18 0,23 0,26 0,26 0,26 0,25 0,28 0,29 0,32 0,33 0,34 0,33 0,31 0,30 0,29 0,31 0,33 0,22 0,30 0,31 0,28
Agg-oilusd (component-specific) 0,13 0,18 0,22 0,25 0,25 0,25 0,24 0,26 0,25 0,26 0,26 0,27 0,26 0,25 0,23 0,20 0,22 0,25 0,21 0,26 0,24 0,24
Agg-multivariate (component-specific) 0,13 0,17 0,21 0,23 0,24 0,24 0,23 0,23 0,22 0,22 0,23 0,23 0,23 0,23 0,22 0,23 0,27 0,31 0,20 0,23 0,25 0,23
Direct forecasts
Indirect forecasts
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 34/37
Tab le.A.8: RMSFE of energy models; c risis and p re-crisis eva luat ion samp le
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 Horizon: 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks Benchmarks
AR(1) 4,43 7,74 8,30 6,83 AR(1) 2,91 4,80 5,53 4,41
ARMA(1,1) 4,45 7,71 8,29 6,82 ARMA(1,1) 2,87 4,76 5,51 4,38
AR(12) 4,86 8,99 9,63 7,83 AR(12) 2,93 4,92 5,77 4,54
Seas. Dummies 4,24 7,62 8,27 6,71 Seas. Dummies 2,82 4,72 5,56 4,37
Univariate (AIC) AR MA Univariate (AIC) AR MA
U.1 9 9 4,70 8,46 8,91 7,36 U.1 6 8 3,54 5,53 5,85 4,97
U.2 2 11 5,47 10,59 10,75 8,94 U.2 8 10 3,02 5,32 6,05 4,80
U.3 4 11 5,43 9,37 9,08 7,96 U.3 8 9 3,41 5,21 6,05 4,89
U.4 9 7 4,92 7,96 8,69 7,19 U.4 10 7 2,95 5,51 5,96 4,81
Univariate (SIC) AR MA Univariate (SIC) AR MA
U.1 5 9 4,77 8,04 8,62 7,15 U.1 6 8 3,54 5,53 5,85 4,97
U.2 1 0 4,14 7,59 8,28 6,67 U.2 2 4 2,93 4,85 5,55 4,44
U.3 2 1 4,43 7,88 8,58 6,96 U.3 6 4 3,06 4,94 5,75 4,58
U.4 1 0 4,04 6,96 7,42 6,14 U.4 1 1 2,89 5,04 5,93 4,62
ADL Specification Y-lags X-lags ADL Specification Y-lags X-lags
oileur B.4 2 12 1,95 3,29 3,62 2,96 oileur B.3 0 12 1,57 2,05 2,24 1,95
oilusd B.4 0 12 1,69 2,40 2,66 2,25 oilusd B.3 0 16 1,67 2,43 2,56 2,22
hwwieur B.2 5 12 1,74 2,67 2,94 2,45 hwwieur B.1 5 12 1,49 1,82 1,89 1,73
hwwicoil B.4 2 12 1,96 3,09 3,38 2,81 hwwicoil B.3 0 14 1,65 1,95 1,99 1,86
hwwienrm B.4 3 12 1,83 2,76 3,01 2,53 hwwienrm B.3 0 14 1,63 1,91 1,91 1,82
hwwi B.2 5 12 1 ,79 2,71 3 ,00 2,50 hwwi B.1 5 12 1 ,46 1,78 1 ,89 1,71
eurusd B.4 1 n.a. 4,04 6,96 7,42 6,14 eurusd B.3 1 n.a. 2,83 4,73 5,57 4,38
iptotencon B.4 1 n.a. 4,04 6,96 7,42 6,14 iptotencon B.3 1 n.a. 2,83 4,73 5,57 4,38
iptotcon B.4 1 1 3,91 5,96 5,66 5,18 iptotcon B.3 1 n.a. 2,83 4,73 5,57 4,38
iptoten B.4 1 1 3,91 6,04 5,72 5,22 iptoten B.3 1 n.a. 2,83 4,73 5,57 4,38
ipcapg B.4 1 n.a. 4,04 6,96 7,42 6,14 ipcapg B.3 1 n.a. 2,83 4,73 5,57 4,38
ipconsg B.4 1 n.a. 4,04 6,96 7,42 6,14 ipconsg B.3 1 n.a. 2,83 4,73 5,57 4,38
ipintgds B.4 1 0 4,28 7,41 7,54 6,41 ipintgds B.3 1 n.a. 2,83 4,73 5,57 4,38
ipener B.4 1 1 3 ,94 6,38 6 ,67 5,66 ipener B.4 11 2 2,84 4,31 4,95 4,03
iocapg B.2 1 7 4,10 5,84 5,01 4,99 iocapg B.1 1 7 2,73 4 ,50 5 ,39 4,21
ioconsd B.4 1 0 4,15 7,24 7,69 6,36 ioconsd B.3 1 0 2,81 4,74 5,58 4,38
iointg B.4 1 2 3,29 4,64 5,17 4,37 iointg B.3 2 2 2,67 4,45 5,18 4,10
uetot_lm B.4 1 n.a. 4,04 6,96 7,42 6,14 uetot_lm B.3 1 n.a. 2,83 4,73 5,57 4,38
uer_lm B.4 1 0 3,87 6,37 6,57 5,60 uer_lm B.3 1 n.a. 2,83 4,73 5,57 4,38
Energy
Crisis Pre-crisis
Tab le.A.9: RMSFE of p roc essed food mod els; c risis and pre-crisis eva luat ion sample
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 Horizon: 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks Benchmarks
AR(1) 0,78 1,88 2,31 1,66 AR(1) 0,52 0,86 0,96 0,78
ARMA(1,1) 0,75 1,83 2,23 1,60 ARMA(1,1) 0,59 1,08 1,39 1,02
AR(12) 0,84 2,24 2,94 2,01 AR(12) 0,53 0,90 1,16 0,86
Seas. Dummies 0,85 1,92 2,31 1,69 Seas. Dummies 0,47 0,83 0,96 0,75
Univariate (AIC) AR MA Univariate (AIC) AR MA
U.1 11 9 0,82 2,17 2,81 1,93 U.1 10 1 0,54 0,88 1,04 0,82
U.2 9 10 0,82 2,03 2,58 1,81 U.2 5 5 0,54 0,99 1,28 0,94
U.3 5 9 0,84 2,10 2,67 1,87 U.3 10 9 0,55 0,93 1,12 0,87
U.4 10 2 0,78 1,92 2,48 1,73 U.4 11 11 0,61 1,11 1,35 1,02
Univariate (SIC) AR MA Univariate (SIC) AR MA
U.1 4 2 0,74 1,83 2,28 1,62 U.1 3 1 0,59 1,04 1,17 0,93
U.2 1 1 0,75 1,91 2,45 1,70 U.2 5 5 0,54 0,99 1,28 0,94
U.3 5 1 0,83 2,17 2,78 1,93 U.3 4 3 0,56 0,86 1,00 0,81
U.4 1 1 0,75 1,94 2,56 1,75 U.4 1 5 0,60 1,18 1,57 1,11
ADL Specification Y-lags X-lags ADL Specification Y-lags X-lags
oileur B.4 10 n.a. 0,78 1,98 2,59 1,78 oileur B.1 9 n.a. 0,54 0,87 0,99 0,80
oilusd B.4 10 n.a. 0,78 1,98 2,59 1,78 oilusd B.1 9 n.a. 0,54 0,87 0,99 0,80
hwwieur B.4 10 n.a. 0,78 1,98 2,59 1,78 hwwieur B.1 9 n.a. 0,54 0,87 0,99 0,80
hwwixeneur B.4 10 0 0,76 1,91 2,53 1,73 hwwixeneur B.1 9 n.a. 0,54 0,87 0,99 0,80
hwwiarm B.4 10 n.a. 0,78 1,98 2,59 1,78 hwwiarm B.1 9 1 0,54 0,90 1,04 0,83
hwwicer B.4 10 0 0,75 1,84 2,41 1,66 hwwicer B.1 9 0 0,54 0,88 1,00 0,81
hwwienrm B.4 10 n.a. 0,78 1,98 2,59 1,78 hwwienrm B.1 9 n.a. 0,54 0,87 0,99 0,80
hwwifood B.4 10 0 0,75 1,87 2,47 1,69 hwwifood B.1 9 0 0,53 0,85 0,97 0,78
hwwiseed B.4 10 0 0,75 1,86 2,45 1,69 hwwiseed B.1 9 2 0,50 0,75 0,86 0,70
hwwi B.4 10 n.a. 0,78 1,98 2,59 1,78 hwwi B.1 9 n.a. 0,54 0,87 0,99 0,80
eurusd B.1 2 1 0,71 1,72 2,15 1,52 eurusd B.3 2 1 0,46 0,73 0,82 0,67
eer21 B.1 10 n.a. 0,78 1,98 2,59 1,78 eer21 B.1 2 1 0,51 0,79 0,86 0,72
reer21 B.1 10 n.a. 0,78 1,98 2,59 1,78 reer21 B.1 2 1 0 ,51 0,78 0 ,84 0,71
iptotencon B.4 10 n.a. 0,78 1,98 2,59 1,78 iptotencon B.1 9 5 0,56 0,88 1,01 0,81
iptotcon B.1 10 0 0,76 1,91 2,50 1,72 iptotcon B.3 9 n.a. 0,50 0,90 1,15 0,85
iptoten B.4 10 0 0,76 1,91 2,50 1,73 iptoten B.1 9 5 0,56 0,89 1,04 0,83
ipcapg B.4 10 n.a. 0,78 1,98 2,59 1,78 ipcapg B.1 9 5 0 ,55 0 ,89 1 ,07 0,84
ipconsg B.1 10 0 0,78 1,96 2,56 1,76 ipconsg B.1 9 5 0,54 0,88 1,03 0,81
ipintgds B.4 10 n.a. 0,78 1,98 2,59 1,78 ipintgds B.1 9 5 0,57 0,89 0,99 0,81
ipener B.2 10 n.a. 0,78 1,98 2,59 1,78 ipener B.4 2 0 0,47 0,74 0,88 0,70
iocapg B.1 9 0 0,75 1 ,94 2,58 1,76 iocapg B.3 9 0 0 ,50 0,90 1 ,16 0,85
ioconsd B.4 10 0 0,78 1,96 2,57 1,77 ioconsd B.1 9 4 0,55 0,90 1,07 0,84
iointg B.3 10 0 0,76 1,93 2,54 1,74 iointg B.1 9 0 0,55 0,88 1,00 0,81
uetot_lm B.3 10 0 0,78 1,96 2,44 1,73 uetot_lm B.4 10 1 0,50 0,89 1,07 0,82
uer_lm B.3 10 0 0,77 1,95 2,49 1,74 uer_lm B.2 2 0 0,46 0,76 0,97 0,73
foodunp B.2 9 0 0,79 2,03 2,62 1,81 foodunp B.1 9 0 0,50 0,81 0,92 0,74
Processed food
Crisis Pre-crisis
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 35/37
33
Tab le A.10: RMSFE of unproc essed food mod els; c risis and p re-crisis evaluation
sample
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 Horizon: 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks Benchmarks
AR(1) 1,21 1,92 1,88 1,67 AR(1) 1,32 1,80 1,63 1,58
ARMA(1,1) 1,20 1,90 1,87 1,66 ARMA(1,1) 1,31 1,77 1,64 1,57
AR(12) 1,07 1,90 2,04 1,67 AR(12) 1,10 1,73 1,78 1,54Seas. Dummies 0,94 1,73 1,86 1,51 Seas. Dummies 0,92 1,51 1,67 1,36
Univariate (AIC) AR MA Univariate (AIC) AR MA
U.1 10 11 1,06 1,83 1,92 1,60 U.1 10 9 1,14 1,76 1,71 1,54
U.2 2 3 0,94 1,73 1,87 1,51 U.2 3 11 1,13 2,07 2,11 1,77
U.3 9 9 1,01 1,82 1,91 1,58 U.3 10 9 1,17 1,78 1,83 1,59
U.4 4 5 0,97 1,72 1,87 1,52 U.4 3 11 1,23 2,27 2,32 1,94
Univariate (SIC) AR MA Univariate (SIC) AR MA
U.1 7 8 1,14 1,83 1,91 1,63 U.1 10 9 1,14 1,76 1,71 1,54
U.2 2 3 0,94 1,73 1,87 1,51 U.2 2 2 0,97 1,56 1,68 1,41
U.3 0 1 1,07 1,95 2,14 1,72 U.3 0 1 1,10 1,81 1,93 1,61
U.4 2 4 0,95 1,75 1,89 1,53 U.4 2 3 0,98 1,52 1,68 1,39
ADL Specification Y-lags X-lags ADL Specification Y-lags X-lags
oileur B.3 1 8 1,14 1,98 2,12 1,75 oileur B.3 0 11 0,76 0,98 1,14 0,96
oilusd B.3 1 8 1,23 2,14 2,29 1,88 oilusd B.3 0 8 0,86 1,23 1,35 1,15
hwwieur B.4 0 8 1,25 2,08 2,19 1,84 hwwieur B.3 0 11 0,77 1,08 1,27 1,04
hwwixeneur B .2 1 8 0,97 1,69 1,76 1,47 hwwixeneur B.3 1 n.a. 0,94 1,52 1,66 1,37
hwwiarm B.2 1 10 1,09 1,76 1,61 1,49 hwwiarm B.4 0 11 0,89 1,52 1,65 1,36
hwwicer B.1 11 1 1,07 1,67 1,71 1,48 hwwicer B.3 1 n.a. 0,94 1,52 1,66 1,37
hwwienrm B.3 0 8 1,20 2,06 2,21 1,82 hwwienrm B.3 0 11 0,74 0,98 1,12 0,95
hwwifood B.4 1 n.a. 0,93 1,71 1,84 1,49 hwwifood B.3 1 n.a. 0,94 1,52 1,66 1,37
hwwiseed B.4 1 8 0,86 1,54 1,68 1,36 hwwiseed B.4 1 5 1,02 1,81 2,02 1,62
hwwiexen B.4 1 n.a. 0,93 1,71 1,84 1,49 hwwiexen B.3 0 8 0,95 1,61 1,82 1,46
hwwi B.4 0 8 1,28 2,13 2,25 1,89 hwwi B.3 0 11 0 ,74 1 ,01 1 ,18 0,98
eurusd B.4 1 n.a. 0,93 1,71 1,84 1,49 eurusd B.3 1 n.a. 0,94 1,52 1,66 1,37
eer21 B.4 1 n.a. 0,93 1,71 1,84 1,49 eer21 B.3 0 11 0,93 1,48 1,54 1,32
reer21 B.4 1 n.a. 0,93 1,71 1,84 1,49 reer21 B.3 1 n.a. 0,94 1,52 1,66 1,37
iptotencon B.2 5 13 0,72 2,05 2,82 1,87 iptotencon B.1 5 14 0,99 1,49 1,66 1,38
iptotcon B.2 5 14 0,78 2,12 3,47 2,12 iptotcon B.1 5 14 0,95 1,45 1,60 1,33
iptoten B.1 5 14 0,86 1,89 3,62 2,12 iptoten B.3 1 18 1,11 1,55 1,69 1,45
ipcapg B.1 5 13 0,72 1,13 1,82 1,22 ipcapg B.2 5 14 1,11 1,56 1,67 1,45
ipconsg B.1 1 10 0,74 1,26 1,42 1,14 ipconsg B.2 1 10 0,94 1,42 1,48 1,28
ipintgds B.3 1 9 1,66 2,71 2,47 2,28 ipintgds B.1 5 14 0,93 1,58 1,81 1,44
ipener B.4 2 1 0 ,93 1,74 1 ,90 1,52 ipener B.2 1 13 1,11 1,82 1,92 1,62
iocapg B.1 5 13 0,72 1,13 1,82 1,22 iocapg B.3 1 1 0 ,92 1 ,46 1 ,62 1,34
ioconsd B.2 5 15 0,93 1,36 1,47 1,25 ioconsd B.2 5 14 1,13 1,44 1,48 1,35
iointg B.4 1 20 1,18 1,82 2,76 1,92 iointg B.1 5 14 0,98 1,54 1,75 1,43
uetot_lm B.3 1 4 0,70 1,05 0,96 0,90 uetot_lm B.3 1 n.a. 0,94 1,52 1,66 1,37
uer_lm B.3 1 5 0,65 1,03 1,04 0,91 uer_lm B.3 1 n.a. 0,94 1,52 1,66 1,37
Unprocessed food
Crisis Pre-crisis
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 36/37
34
Tab le A.11: RMSFE of non-ene rgy ind ustria l good s models; c risis and pre-crisis
evaluation sample
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 Horizon: 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks Benchmarks
AR(1) 1,53 1,62 0,33 1,16 AR(1) 1,22 1,31 0,40 0,98
ARMA(1,1) 1,51 1,60 0,33 1,15 ARMA(1,1) 1,20 1,29 0,38 0,96
AR(12) 0,17 0,28 0,32 0,26 AR(12) 0,18 0,34 0,39 0,30Seas. Dummies 0,64 0,70 0,30 0,54 Seas. Dummies 0,55 0,60 0,34 0,50
Univariate (AIC) AR MA Univariate (AIC) AR MA
U.1 11 11 0,18 0,29 0,30 0,26 U.1 11 8 0,25 0,55 0,60 0,47
U.2 8 5 0,17 0,28 0,30 0,25 U.2 10 11 0,21 0,37 0,38 0,32
U.3 7 10 0,19 0,27 0,30 0,25 U.3 9 11 0,18 0,35 0,41 0,31
U.4 11 7 0,19 0,28 0,30 0,26 U.4 9 10 0,17 0,32 0,36 0,28
Univariate (SIC) AR MA Univariate (SIC) AR MA
U.1 11 5 0,19 0,29 0,30 0,26 U.1 11 8 0,25 0,55 0,60 0,47
U.2 5 5 0,15 0,26 0,29 0,24 U.2 5 5 0,17 0,31 0,35 0,28
U.3 6 0 0,16 0,29 0,33 0,26 U.3 7 7 0,17 0,30 0,34 0,27
U.4 6 0 0,14 0,27 0,32 0,24 U.4 5 7 0,16 0,31 0,36 0,28
ADL Specification Y-lags X-lags ADL Specification Y-lags X-lags
oileur B.4 6 0 0,13 0,26 0,31 0,23 oileur B.1 6 5 0,17 0,26 0,27 0,23
oilusd B.4 6 0 0,13 0,26 0,32 0,24 oilusd B.1 6 5 0,16 0,25 0,27 0,23
hwwieur B.4 6 0 0,13 0,26 0,31 0,23 hwwieur B.1 11 5 0,18 0,28 0,28 0,25
hwwiarm B.3 11 1 0,16 0,26 0,28 0,23 hwwiarm B.1 11 5 0,19 0,34 0,39 0,30
hwwicoal B.4 6 n.a. 0,14 0,27 0,32 0,24 hwwicoal B.1 6 1 0,16 0,30 0,35 0,27
hwwicoil B.3 11 0 0,16 0,25 0,28 0,23 hwwicoil B.1 6 5 0,17 0,26 0,26 0,23
hwwienrm B.3 11 0 0,16 0,26 0,28 0,23 hwwienrm B.1 6 5 0,17 0,27 0,26 0,23
hwwiind B.3 11 n.a. 0,17 0,27 0,29 0,24 hwwiind B.4 6 n.a. 0,17 0,32 0,37 0,29
hwwiiost B.3 11 n.a. 0,17 0,27 0,29 0,24 hwwiiost B.3 11 4 0,20 0,31 0,31 0,27
hwwinfm B.4 6 n.a. 0,14 0,27 0,32 0,24 hwwinfm B.4 6 n.a. 0,17 0,32 0,37 0,29
hwwitex B.3 11 10 0,18 0,26 0,29 0,25 hwwitex B.1 6 1 0,17 0,31 0,35 0,27
hwwiexen B.3 11 n.a. 0,17 0,27 0,29 0,24 hwwiexen B.4 6 n.a. 0,17 0,32 0,37 0,29
hwwi B.3 11 0 0 ,16 0,25 0 ,28 0,23 hwwi B.1 11 5 0 ,18 0,28 0 ,28 0,25
eurusd B.4 11 16 0,17 0,28 0,32 0,26 eurusd B.3 11 4 0,20 0,36 0,40 0,32
eer21 B.2 11 24 0,18 0,24 0,25 0,22 eer21 B.3 11 4 0 ,19 0,35 0,39 0,31
eer41 B.3 11 5 0,18 0,27 0,29 0,25 eer41 B.3 11 4 0,18 0,33 0,36 0,29
reer21 B.3 11 5 0,18 0,27 0,29 0,24 reer21 B.3 11 4 0,20 0,36 0,40 0,32
reer41 B.4 6 n.a. 0,14 0,27 0,32 0,24 reer41 B.3 11 4 0,19 0,35 0,38 0,31
iptotencon B.4 6 n.a. 0,14 0,27 0,32 0,24 iptotencon B.1 6 4 0,16 0,31 0,36 0,28
iptotcon B.3 11 n.a. 0,17 0,27 0,29 0,24 iptotcon B.1 6 4 0,16 0,30 0,35 0,27
iptoten B.3 11 n.a. 0,17 0,27 0,29 0,24 iptoten B.2 6 4 0,17 0,32 0,39 0,29
ipcapg B.4 6 n.a. 0,14 0,27 0,32 0,24 ipcapg B.4 6 n.a. 0,17 0,32 0,37 0,29
ipconsg B.1 11 9 0,17 0,22 0,21 0,20 ipconsg B.1 6 4 0 ,16 0,31 0,36 0,27
ipintgds B.3 11 n.a. 0,17 0,27 0,29 0,24 ipintgds B.1 6 4 0,17 0,31 0,36 0,28
ipener B.1 11 5 0,17 0,27 0,29 0,24 ipener B.3 7 n.a. 0,16 0,31 0,36 0,28
iocapg B.3 11 n.a. 0,17 0,27 0,29 0,24 iocapg B.1 6 4 0 ,18 0 ,31 0 ,36 0,28
ioconsd B.3 11 n.a. 0,17 0,27 0,29 0,24 ioconsd B.3 6 8 0,22 0,38 0,42 0,34
iointg B.2 6 1 0,16 0,29 0,33 0,26 iointg B.1 6 4 0,17 0,31 0,35 0,27
hhfinsit12 B.3 11 n.a. 0,17 0,27 0,29 0,24 hhfinsit12 B.1 6 0 0,17 0,31 0,36 0,28
majpur12 B.4 6 n.a. 0,14 0,27 0,32 0,24 majpur12 B.3 11 3 0,17 0,30 0,34 0,27
pritre12 B.4 6 0 0,14 0,27 0,32 0,24 pritre12 B.1 6 n.a. 0,17 0,31 0,36 0,28
sav12 B.3 11 n.a. 0,17 0,27 0,29 0,24 sav12 B.3 11 0 0,17 0,31 0,35 0,28
uetot_lm B.4 6 0 0,13 0,24 0,25 0,21 uetot_lm B.3 7 0 0,16 0,29 0,33 0,26
uer_lm B.4 6 0 0,13 0,24 0,24 0,21 uer_lm B.4 6 n.a. 0,17 0,32 0,37 0,29
Non-energy industrial goods
Crisis Pre-crisis
7/30/2019 Impactul crizei asupra modelelor econometrice
http://slidepdf.com/reader/full/impactul-crizei-asupra-modelelor-econometrice 37/37
Tab le A.12: RMSFE of services models; crisis and pre-c risis eva lua tion sample
Horizon: 1 to 6 7 to 12 13 to 18 1 to 18 Horizon: 1 to 6 7 to 12 13 to 18 1 to 18
Benchmarks Benchmarks
AR(1) 0,50 0,56 0,51 0,52 AR(1) 0,45 0,46 0,28 0,40
ARMA(1,1) 0,44 0,50 0,50 0,48 ARMA(1,1) 0,40 0,43 0,32 0,38
AR(12) 0,23 0,41 0,56 0,40 AR(12) 0,20 0,34 0,40 0,31
Seas. Dummies 0,18 0,36 0,50 0,35 Seas. Dummies 0,16 0,25 0,30 0,23
Univariate (AIC) AR MA Univariate (AIC) AR MA
U.1 9 11 0,26 0,41 0,52 0,40 U.1 11 7 0,24 0,32 0,33 0,30
U.2 10 6 0,18 0,37 0,51 0,35 U.2 10 8 0,15 0,25 0,31 0,23
U.3 11 11 0,23 0,41 0,54 0,39 U.3 9 9 0,18 0,26 0,31 0,25
U.4 10 9 0,20 0,40 0,53 0,38 U.4 11 5 0,13 0,23 0,30 0,22
Univariate (SIC) AR MA Univariate (SIC) AR MA
U.1 9 11 0,26 0,41 0,52 0,40 U.1 11 7 0,24 0,32 0,33 0,30
U.2 10 6 0,18 0,37 0,51 0,35 U.2 10 6 0,15 0,25 0,30 0,23
U.3 0 4 0,24 0,45 0,64 0,44 U.3 0 4 0,18 0,35 0,47 0,33
U.4 5 6 0,20 0,38 0,52 0,37 U.4 5 5 0,12 0,21 0,28 0,20
ADL Specification Y-lags X-lags ADL Specification Y-lags X-lags
oileur B.3 11 0 0,20 0,38 0,51 0,36 oileur B.4 7 0 0,15 0,26 0,30 0,24
oilusd B.3 11 0 0,20 0,39 0,52 0,37 oilusd B.4 7 0 0,15 0,26 0,32 0,24
hwwieur B.3 11 2 0,20 0,37 0,50 0,36 hwwieur B.4 7 0 0,15 0,27 0,31 0,24
hwwixeneur B.3 11 n.a. 0,19 0,36 0,49 0,35 hwwixeneur B.3 11 n.a. 0,15 0,27 0,35 0,26
hwwiarm B.3 11 0 0,19 0,36 0,47 0,34 hwwiarm B.3 11 n.a. 0,15 0,27 0,35 0,26
hwwicoil B.3 11 2 0,20 0,38 0,50 0,36 hwwicoil B.4 7 0 0,15 0,26 0,30 0,23
hwwienrm B.3 11 2 0,20 0,38 0,51 0,36 hwwienrm B.4 7 0 0,15 0,26 0,30 0,24
hwwifood B.3 11 n.a. 0,19 0,36 0,49 0,35 hwwifood B.3 11 1 0,15 0,28 0,36 0,26
hwwiind B.3 11 n.a. 0,19 0,36 0,49 0,35 hwwiind B.3 11 n.a. 0,15 0,27 0,35 0,26
hwwinfm B.3 11 n.a. 0,19 0,36 0,49 0,35 hwwinfm B.3 11 n.a. 0,15 0,27 0,35 0,26
hwwiseed B.3 11 n.a. 0,19 0,36 0,49 0,35 hwwiseed B.4 7 1 0,15 0,29 0,35 0,26
hwwitex B.3 11 n.a. 0,19 0,36 0,49 0,35 hwwitex B.3 11 n.a. 0,15 0,27 0,35 0,26
hwwiexen B.3 11 n.a. 0,19 0,36 0,49 0,35 hwwiexen B.3 11 n.a. 0,15 0,27 0,35 0,26
hwwi B.3 11 2 0 ,20 0,38 0 ,51 0,37 hwwi B.4 7 2 0,15 0,28 0,33 0,25
eurusd B.3 11 0 0,18 0,34 0,46 0,33 eurusd B.3 11 n.a. 0,15 0,27 0,35 0,26
eer21 B.3 11 0 0,18 0,34 0,45 0,32 eer21 B.3 11 0 0,14 0,23 0,26 0,21
eer41 B.3 11 0 0,18 0,34 0,46 0,33 eer41 B.3 11 0 0,14 0,24 0,30 0,23
reer21 B.3 11 0 0,18 0,34 0,45 0,32 reer21 B.3 11 0 0,14 0,23 0,26 0,21
reer41 B.3 11 0 0,18 0,34 0,45 0,32 reer41 B.3 11 0 0,14 0,24 0,28 0,22
iptotencon B.2 11 15 0,31 0,52 0,60 0,48 iptotencon B.3 6 5 0,15 0,27 0,33 0,25
iptotcon B.4 5 12 0,27 0,68 0,97 0,64 iptotcon B.1 11 11 0,16 0,25 0,33 0,25
iptoten B.3 11 n.a. 0,19 0,36 0,49 0,35 iptoten B.1 7 15 0,17 0,24 0,34 0,25
ipcapg B.2 9 15 0,30 0,57 0,71 0,53 ipcapg B.1 11 11 0,19 0,26 0,33 0,26
ipconsg B.3 8 5 0,18 0,32 0,45 0,32 ipconsg B.3 6 5 0,15 0,25 0,31 0,24
ipintgds B.2 8 15 0,31 0,57 0,64 0,51 ipintgds B.3 7 5 0,15 0,26 0,33 0,25
ipener B.4 7 n.a. 0,18 0,38 0,54 0,37 ipener B.2 11 17 0,15 0,25 0,32 0,24
iocapg B.4 7 0 0,19 0,40 0,57 0,39 iocapg B.4 7 0 0 ,16 0 ,29 0 ,34 0,26
ioconsd B.3 11 1 0,20 0,41 0,55 0,39 ioconsd B.3 11 5 0,18 0,28 0,29 0,25
iointg B.2 5 24 0,29 0,45 0,47 0,40 iointg B.3 7 12 0,19 0,31 0,39 0,30
uetot_lm B.4 8 17 0,26 0,53 0,69 0,49 uetot_lm B.3 11 n.a. 0,15 0,27 0,35 0,26
uer_lm B.4 7 10 0,22 0,34 0,41 0,32 uer_lm B.3 11 n.a. 0,15 0,27 0,35 0,26
Services
Crisis Pre-crisis